[{"_1":2,"_291":-5,"_292":-5},"loaderData",{"_3":4,"_8":9,"_23":24},"root",{"_5":-5,"_6":7},"user","countable",false,"routes/_main",{"_5":-5,"_10":11},"recentPosts",[12,17,20],{"_13":14,"_15":16},"slug","claude-code-gitlab-support","title","Claude Code's GitLab Support: What Changed",{"_13":18,"_15":19},"github-copilot-pricing-tiers","GitHub Copilot Pricing: Which Tier Do You Actually Need?",{"_13":21,"_15":22},"claude-code-self-hosted-runner-production","Deploying a Self-Hosted Claude Code Runner to Production","routes/creators.$handle",{"_25":26,"_27":26,"_28":29,"_30":7,"_31":32,"_289":7,"_290":110},"handle","kangarooking","displayName","profileUrl","https://github.com/kangarooking","verified","skills",[33],{"_34":35,"_13":36,"_37":38,"_39":40,"_41":42,"_43":44,"_45":46,"_47":48,"_49":50,"_51":52,"_53":54,"_59":60,"_61":62,"_63":64,"_80":81,"_82":-5,"_83":84,"_85":86,"_87":88,"_89":90,"_91":92,"_93":-5,"_94":95,"_96":26,"_30":7,"_97":7,"_98":86,"_99":100,"_101":86,"_102":86,"_103":104,"_105":106,"_134":135,"_170":171,"_221":222,"_263":264,"_275":276,"_285":286,"_287":288},"id","gh__kangarooking__cangjie-skill__root","cangjie-skill-kangarooking","name","Cangjie Skill","canonicalName","cangjie-skill","tagline","Transform long content into actionable skills.","description","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).","longDescription","Cangjie Skill is designed to distill extensive content such as books, video transcripts, podcasts, and courses into a coherent set of executable skills. This skill is particularly useful for users who want to extract frameworks, principles, and methodologies from long-form materials and convert them into atomic skills that can be invoked in real-world situations. Unlike simple summarization or book reviews, Cangjie focuses on creating reusable skills that can directly assist users in applying the knowledge gleaned from the original content.\n\nThe process involves a structured methodology known as RIA-TV++, which includes five stages of extraction and validation. Initially, the skill requires the user to provide a text source, which can be in various formats like PDF or TXT. The skill then reads the content, extracting key methodologies and frameworks through a series of parallel tasks performed by sub-agents. Each extracted unit undergoes a triple verification process to ensure its validity and uniqueness before being transformed into a skill format.\n\nOnce the skills are constructed, they are linked together using Zettelkasten principles, creating an interconnected knowledge base. This facilitates easy navigation and reference among related skills. The final output includes a digest of the distilled content, allowing users to access essential insights without having to read the entire source material. Additionally, the skills are designed to be compatible with darwin-skill for future evolution and improvement.\n\nCangjie Skill is ideal for developers, educators, and lifelong learners who seek to leverage extensive knowledge from various sources into practical, actionable skills. It empowers users to efficiently utilize information and apply it in their specific contexts, enhancing productivity and learning outcomes.","primaryCategory","ai-agents","subcategory","skill-authoring","secondaryCategories",[],"tags",[50,55,56,57,58],"methodology-extraction","content-distillation","frameworks","principles","skillType","workflow","complexity","intermediate","platforms",[65,66,67,68,69,70,71,72,73,74,75,76,77,78,79],"claude-code","claude-desktop","claude-api","codex-cli","cursor","antigravity","github-copilot","windsurf","opencode","cline","roo-code","goose","kiro","continue","trae","delivery","source-link","skillFileUrl","pricing","free","price",0,"thumbnailType","icon","thumbnailIcon","lightbulb","thumbnailColor","blue","thumbnailImageUrl","sellerId","gh__kangarooking","sellerName","featured","downloads","views",4,"rating","reviewCount","status","active","metrics",{"_107":108,"_109":110,"_111":110,"_112":108,"_113":114,"_115":116,"_117":118,"_119":120,"_121":122,"_123":124,"_125":126,"_127":128,"_129":110,"_130":131,"_132":133},"freshnessScore",100,"mirrorCount",1,"skillCountInRepo","qualityScore","popularityScore",79,"trustScore",60,"repoOpenIssues",12,"repoStars",7023,"repoWatchers",29,"repoContributors",5,"overallScore",87,"repoCreatedAt","2026-04-16T05:54:32Z","daysSinceUpdate","repoForks",882,"repoPushedAt","2026-08-09T13:34:40Z","source",{"_136":26,"_137":138,"_139":140,"_145":40,"_146":-5,"_143":147,"_148":149,"_150":7,"_151":152,"_153":154,"_160":161,"_162":7,"_163":164,"_165":166,"_167":168,"_169":168},"owner","repoUrl","https://github.com/kangarooking/cangjie-skill","author",{"_28":29,"_141":142,"_37":26,"_25":26,"_143":144},"avatarUrl","https://github.com/kangarooking.png","type","User","repo","commitSha","github","branch","main","isOfficial","path","","license",{"_37":155,"_156":157,"_158":159},"GNU Affero General Public License v3.0","spdxId","AGPL-3.0","url","https://github.com/kangarooking/cangjie-skill/blob/main/LICENSE","sourceUpdatedAt","2026-08-07T04:46:10.000Z","redistributable","permalink","https://github.com/kangarooking/cangjie-skill/blob/main/SKILL.md","repoFullName","kangarooking/cangjie-skill","lastSyncedAt","2026-08-11T01:22:52.641Z","firstSeenAt","seo",{"_172":173,"_184":185,"_186":187,"_205":-5,"_206":207,"_213":214,"_215":216,"_217":218,"_219":220},"useCases",[174,178,181],{"_15":175,"_176":177},"Distilling a Business Book","body","You have a business book filled with strategies and frameworks. Use Cangjie Skill to extract these into actionable skills for your team.",{"_15":179,"_176":180},"Creating Skills from Educational Videos","After watching a long educational video, you want to apply its concepts. Cangjie Skill can distill the video into skills that you can use.",{"_15":182,"_176":183},"Extracting Principles from a Podcast","You listened to a podcast that discussed various methodologies. Use this skill to transform those discussions into practical skills.","whenNotToUse","This skill is not suitable for simple summarization or when the user does not have a text source available for extraction.","faqs",[188,193,196,199,202],{"_189":190,"_191":192},"q","What types of content can I distill using Cangjie Skill?","a","You can distill various forms of long content, including books, video transcripts, podcasts, and courses.",{"_189":194,"_191":195},"How does the extraction process work?","The skill uses a structured methodology involving multiple extraction phases and validation steps to ensure the quality of the skills created.",{"_189":197,"_191":198},"Can I use Cangjie Skill for simple summaries?","No, Cangjie Skill is specifically designed for creating actionable skills, not for simple summarization or reviews.",{"_189":200,"_191":201},"What happens if I don't have a text source?","The skill requires a text source to function. If none is provided, it will prompt you to supply one before proceeding.",{"_189":203,"_191":204},"Is it possible to batch process multiple contents?","While you can batch process, it is recommended to first pilot with one piece of content to ensure the process works smoothly.","ogImageUrl","keywords",[208,209,210,211,212],"distill book into skills","extract methodologies","convert video to skills","long content to skills","actionable insights","whenToUse","Use Cangjie Skill when you want to distill a book, video, or podcast into practical skills that can be executed by AI agents.","metaTitle","Cangjie Skill - Transform Long Content into Actionable Skills","problemSolved","This skill removes the difficulty of extracting actionable insights from lengthy content, enabling users to apply learned methodologies directly.","metaDescription","Cangjie Skill distills long content into actionable skills for practical use. Discover how it works!","content",{"_223":124,"_224":225,"_248":249,"_256":257,"_258":7,"_259":260,"_261":262},"bundledFiles","fileTree",[226,233,236,239,242,245],{"_151":227,"_228":229,"_143":230,"_231":232},"GITHUB_REPO.md","size",454,"file","language","markdown",{"_151":234,"_228":235,"_143":230},"LICENSE",34523,{"_151":237,"_228":238,"_143":230,"_231":232},"README.en.md",14645,{"_151":240,"_228":241,"_143":230,"_231":232},"README.ja.md",16449,{"_151":243,"_228":244,"_143":230,"_231":232},"README.md",16972,{"_151":246,"_228":247,"_143":230,"_231":232},"SKILL.md",9956,"codeExamples",[250,254],{"_231":251,"_252":253},"text","snippet","阶段 0: Adler 整书理解 → BOOK_OVERVIEW.md\n阶段 1: 5 个 agent 并行提取 → 候选方法论单元池\n阶段 1.5: 三重验证筛选 → 通过的单元 (用户轻确认)\n阶段 2: RIA++ 构造 skill → 每个 skill 的 SKILL.md\n阶段 3: Zettelkasten 链接 → INDEX.md + GLOSSARY.md\n阶段 4: 压力测试 (darwin 兼容) → test-prompts.json + 回炉淘汰\n阶段 5: 交付 → DIGEST.md 精华长文 + 安装到 skills 目录",{"_231":251,"_252":255},"books//\n├── PIPELINE_STATE.md # 流水线状态: 当前阶段 + 各 skill 进度 (断点续跑用)\n├── BOOK_OVERVIEW.md # 阶段 0 产出: 主旨/骨架/术语/批判\n├── verified.md # 阶段 1.5 产出: 通过三重验证的单元 + 判定理由\n├── INDEX.md # 阶段 3 产出: skill 总览 + 引用图\n├── GLOSSARY.md # 阶段 3 产出: 全书共享术语词典\n├── DIGEST.md # 阶段 5 产出: 面向读者的精华长文\n├── candidates/ # 阶段 1 产出: 原始候选池 (审计用)\n├── rejected/ # 阶段 1.5 淘汰的单元 + 原因 (审计用)\n├── /\n│ ├── SKILL.md\n│ ├── test-prompts.json # darwin-skill 兼容格式\n│ └── test-results.md # 阶段 4 测试通过率 + 失败分析\n├── /\n│ └── ...","skillMd","# cangjie-skill — 把一本书蒸馏成一组可执行 skills 的元 skill\n\n## 使命\n\n把一本书里沉淀的方法论,拆解成一组**原子化、可被 agent 在真实场景下调用**的 skills,让读者真正用起来。\n\n> **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。\n\n**边界**:\n- ✅ 做: 方法论 / 决策框架 / 清单 / 原则 / 概念体系的蒸馏\n- ❌ 不做: 书摘 / 读后感 / 作者人设角色扮演 (后者请用 nuwa-skill)\n\n## 核心方法论: RIA-TV++\n\n一个五阶段 + 并行提取 + 三重验证 + darwin 兼容测试的流水线。详见 `methodology/00-overview.md`。\n\n```\n阶段 0: Adler 整书理解 → BOOK_OVERVIEW.md\n阶段 1: 5 个 agent 并行提取 → 候选方法论单元池\n阶段 1.5: 三重验证筛选 → 通过的单元 (用户轻确认)\n阶段 2: RIA++ 构造 skill → 每个 skill 的 SKILL.md\n阶段 3: Zettelkasten 链接 → INDEX.md + GLOSSARY.md\n阶段 4: 压力测试 (darwin 兼容) → test-prompts.json + 回炉淘汰\n阶段 5: 交付 → DIGEST.md 精华长文 + 安装到 skills 目录\n```\n\n## 何时调用此 skill\n\n用户说类似:\n- \"帮我拆《穷查理宝典》\"\n- \"把毛选蒸馏成 skill\"\n- \"把这个 B 站视频/播客/课程蒸馏成 skill\"\n- \"distill this book into skills: \"\n- \"我想把这本书的方法论做成可用的 skill\"\n\n## 输入要求\n\n在开始前**必须**从用户处确认:\n1. **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)\n2. **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。\n3. **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。\n\n**非书籍内容的字段映射**: `source_chapter` 等\"章节\"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。\n\n## 输出结构\n\n```\nbooks//\n├── PIPELINE_STATE.md # 流水线状态: 当前阶段 + 各 skill 进度 (断点续跑用)\n├── BOOK_OVERVIEW.md # 阶段 0 产出: 主旨/骨架/术语/批判\n├── verified.md # 阶段 1.5 产出: 通过三重验证的单元 + 判定理由\n├── INDEX.md # 阶段 3 产出: skill 总览 + 引用图\n├── GLOSSARY.md # 阶段 3 产出: 全书共享术语词典\n├── DIGEST.md # 阶段 5 产出: 面向读者的精华长文\n├── candidates/ # 阶段 1 产出: 原始候选池 (审计用)\n├── rejected/ # 阶段 1.5 淘汰的单元 + 原因 (审计用)\n├── /\n│ ├── SKILL.md\n│ ├── test-prompts.json # darwin-skill 兼容格式\n│ └── test-results.md # 阶段 4 测试通过率 + 失败分析\n├── /\n│ └── ...\n```\n\n## 执行流程 (严格按顺序)\n\n**断点续跑**: 开始前先检查 `books//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 (当前阶段 / 已完成产物 / 各 skill 状态 / 下一步),格式用简单的 checklist markdown 即可。\n\n### 阶段 0 — 整书理解\n\n1. 读取用户提供的书本文本。大文件分块阅读。\n2. 执行 `methodology/01-stage0-adler.md` 中的 Adler 四步 (结构 / 解释 / 批判 / 应用)。\n3. 按 `templates/BOOK_OVERVIEW.md.template` 填充,写入 `books//BOOK_OVERVIEW.md`。\n4. 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。\n\n### 阶段 1 — 5 个 sub-agent 并行提取\n\n**并行** spawn 5 个 Task sub-agents(使用 Agent 工具,一次调用中发起 5 个):\n\n| sub-agent | 读取的 prompt | 产出 |\n|---|---|---|\n| 框架提取器 | `extractors/framework-extractor.md` | 决策框架 / 思维模型 |\n| 原则提取器 | `extractors/principle-extractor.md` | 原则 / 清单 / 规则 |\n| 案例提取器 | `extractors/case-extractor.md` | 作者在书中亲自使用过的实例 |\n| 反例提取器 | `extractors/counter-example-extractor.md` | 书中警告的失败模式 |\n| 术语提取器 | `extractors/glossary-extractor.md` | 关键概念词典 |\n\n每个 sub-agent 独立读书、独立提取、独立输出到 `books//candidates/.md`。\n\n- **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。\n- **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。\n\n### 阶段 1.5 — 三重验证筛选\n\n读取 `methodology/03-stage1.5-triple-verify.md`,对每个候选单元执行:\n\n- **V1 跨域**: 书中至少 2 个独立段落有佐证?\n- **V2 预测力**: 能用它回答一个书里没明说的新问题吗?\n- **V3 独特性**: 不是任何聪明人都会说的常识吗?\n\n通过的写入 `books//verified.md`。不通过的写入 `books//rejected/` 并附原因 — 保留审计轨迹,也允许用户事后捞回。\n\n**用户轻确认** ★: 筛选完成后,把\"通过的 N 个候选标题 + 淘汰的 M 个\"列表展示给用户:\"这 N 个会做成 skill,有想捞回或砍掉的吗?\" 得到确认再进入阶段 2 — 阶段 2–4 是最耗时的部分,这一步确认能避免大量返工。\n\n### 阶段 2 — RIA++ 构造 skill\n\n对每个通过的单元,按 `templates/SKILL.md.template` 填充:\n\n- **R (Reading)**: 原文引用 ≤150 字/段 (英文原文 ≤100 词/段)\n- **I (Interpretation)**: 用自己的话重写方法论骨架 (避免照搬译本)\n- **A1 (Past Application)**: 书中作者用过的案例\n- **A2 (Future Trigger)** ★: 用户在什么情境下会需要这个 → skill 的 `description` 字段\n- **E (Execution)**: 1-2-3 可执行步骤\n- **B (Boundary)**: 什么时候不适用 / 来自阶段 0 批判阶段的作者盲点\n\n细则见 `methodology/04-stage2-ria-plus.md`。注意: A2 中\"与相邻 skill 的区分\"此时只写**初稿** (基于 verified.md 的单元列表),阶段 3 建立链接后回填定稿。\n\n### 阶段 3 — Zettelkasten 链接\n\n按 `methodology/05-stage3-zettelkasten.md`:\n1. 找出 skill 之间的引用关系 (A 依赖 B / A 对比 B / A 组合 B)\n2. 在每个 SKILL.md 末尾补\"相关 skills\"段,并回填 A2 的\"与相邻 skill 的区分\"\n3. 按 `templates/INDEX.md.template` 生成 `INDEX.md` (含引用图 mermaid)\n4. 把 `candidates/glossary.md` 整理成 `books//GLOSSARY.md` — 它是所有 skill 的共享词典,不该埋在审计目录里\n\n### 阶段 4 — 压力测试 (darwin 兼容)\n\n对每个 skill 按 `methodology/06-stage4-pressure-test.md`:\n1. 设计 5–10 条测试 prompt,按 `templates/test-prompts.json.template` 写入 `test-prompts.json`\n2. 至少包括 3 类: **应调用** / **不应调用 (诱饵)** / **边界模糊**。诱饵中至少 1 条必须是\"应触发同书另一个 skill\"的场景 (跨 skill 混淆测试)\n3. 优先用独立 sub-agent 盲测每条 prompt,由主流程对照预期统计结果,**未过的回炉重做阶段 2** — 不做\"表面修补\"\n4. 每个 skill 的测试结果写入 `/test-results.md`\n\n### 阶段 5 — 交付\n\n按 `methodology/07-stage5-deliver.md`:\n1. 生成 `books//DIGEST.md` — 面向读者的精华长文 (按 `templates/DIGEST.md.template`),满足\"不读全书、只看精华\"的需求\n2. 询问用户安装位置 (用户级 `~/.claude/skills/` 或项目级 `.claude/skills/` / `.cursor/skills/`),把通过测试的 skill 复制或 symlink 过去 — **没有这一步,产出的 skill 无法被真正调用**\n3. 告知用户: \"已完成,可一键喂给 darwin-skill 自动进化\"\n\n## 质量红线 (违反则阻止输出)\n\n1. 每个 skill 必须通过**全部**三重验证\n2. 每个 skill 必须有完整的 R / I / A1 / A2 / E / B 六段\n3. 原文引用 ≤150 字/段 (英文 ≤100 词/段)\n4. 每个 skill 必须有 `test-prompts.json`,且包含诱饵测试 (不应调用的场景),其中至少 1 条是同书兄弟 skill 的场景\n5. `description` 字段必须明确 trigger 条件,不能只是\"一个关于 X 的 skill\"\n\n## 与 nuwa-skill / darwin-skill 的生态定位\n\n- **nuwa-skill**: 蒸馏人 (思维方式 / 表达 DNA)\n- **cangjie-skill** (本 skill): 蒸馏书 (方法论 / 框架 / 原则)\n- **darwin-skill**: 进化任意 skill\n\n三者咬合: 本 skill 输出的 `test-prompts.json` 严格遵循 darwin-skill 格式,以便产出的 skill 可直接接入 darwin 做自动进化。\n\n## 调用惯例\n\n- **永远先试点 1 本** — 除非用户明确说\"批量\"\n- **阶段之间主动汇报进度** — 不要静默跑完再 dump 结果\n- **不凭记忆拆书** — 没文本就停下来问\n- **保留审计轨迹** — candidates/ 和 rejected/ 都要留\n- **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","skillMdTruncated","frontmatter",{"_37":40,"_43":44},"totalSizeBytes",92999,"runtime",{"_265":7,"_266":7,"_267":-5,"_268":7,"_269":7,"_270":7,"_271":272,"_273":274},"requiresBash","requiresMcp","model","requiresNode","requiresNetwork","requiresPython","allowedTools",[],"dependencies",[],"install",{"_277":278,"_279":166,"_280":138,"_281":282,"_283":284},"cli","npx skills add kangarooking/cangjie-skill","coordinate","browseUrl","downloadUrl","https://github.com/kangarooking/cangjie-skill/archive/refs/heads/main.zip","gitUrl","https://github.com/kangarooking/cangjie-skill.git","createdAt","2026-08-11T01:22:57.811Z","updatedAt","2026-08-15T01:22:07.311Z","hasNext","page","actionData","errors"]