メニュー
Karpathy 式持久 wiki（llm-wiki/）— 按门类路径组织 raw/wiki，Collect 先清单后落盘，ingest/query/lint/graph；融合 llm-wiki-agent、llm-wiki-skill、graphify 的设计思路，工具无 API Key。知识库根目录为仓库内 llm-wiki/。
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
| name | llm-wiki |
| description | Karpathy 式持久 wiki（llm-wiki/）— 按门类路径组织 raw/wiki，Collect 先清单后落盘，ingest/query/lint/graph；融合 llm-wiki-agent、llm-wiki-skill、graphify 的设计思路，工具无 API Key。知识库根目录为仓库内 llm-wiki/。 |
llm-wiki 工作流依赖以下外部 skill。Agent 在执行任何操作前，必须强制检查对应 skill 是否已安装在当前环境中。
检查方式：验证 skill 是否存在于 agent skill 目录（如 .agents/skills/<name>/、.qoder/skills/<name>/）或技能锁文件（skills-lock.json）中。
| 操作阶段 | 依赖 Skill | 安装命令 |
|---|---|---|
| Collect | tavily-search | npx skills add https://github.com/tavily-ai/skills --skill tavily-search |
| Collect | baoyu-url-to-markdown | npx skills add https://github.com/jimliu/baoyu-skills --skill baoyu-url-to-markdown |
| Collect | wechat-article-to-markdown | npx skills add https://github.com/jackwener/wechat-article-to-markdown --skill wechat-article-to-markdown |
| Collect | baoyu-youtube-transcript | npx skills add https://github.com/jimliu/baoyu-skills --skill baoyu-youtube-transcript |
| Ingest（成稿） | humanizer-zh | npx skills add https://github.com/op7418 --skill humanizer-zh |
| Ingest（可视化） | fireworks-tech-graph | npx skills add https://github.com/yizhiyanhua-ai/fireworks-tech-graph --skill fireworks-tech-graph |
| Ingest（演示） | html-ppt-skill | npx skills add https://skills.sh/lewislulu/html-ppt-skill/html-ppt |
强制规则：
Do not skip this check. Do not proceed without user confirmation.
llm-wiki/（与仓库根相对路径；也可单独拷贝本目录对外开源）
llm-wiki/CLAUDE.mdllm-wiki/taxonomy.mdllm-wiki/wiki/index.md 与 wiki/overview.md按任务追加：Collect → llm-wiki/skills/collect/SKILL.md + llm-wiki/tools/source_registry.tsv；写成稿 → llm-wiki/skills/wiki-writing/SKILL.md + llm-wiki/skills/humanizer-zh/SKILL.md
export WIKI_ROOT="$PWD/llm-wiki"
pip install -r llm-wiki/requirements.txt
bash llm-wiki/tools/check_all.sh "$WIKI_ROOT"
python3 llm-wiki/tools/build_graph.py --wiki-root "$WIKI_ROOT"
tools/source_registry.tsvsessions/.../candidates.md → 用户确认 → 写入 raw/<门类>/… → ingestraw/<shelf>/… 与 wiki/<shelf>/… 对齐Fetch any URL and convert to markdown using baoyu-fetch CLI (Chrome CDP with site-specific adapters). Built-in adapters for X/Twitter, YouTube transcripts, Hacker News threads, and generic pages via Defuddle. Handles login/CAPTCHA via interaction wait modes. Use when user wants to save a webpage as markdown.
E2E testing for local web applications — constraint-first, execution-driven structure: iron rules, 5-stage execution flow with inline principles, artifact specification. Use this skill whenever the user wants to test a local web app, validate UI behavior, walk through multi-step user flows, mock API responses for testing, debug E2E test failures, capture screenshots, or understand E2E testing best practices. Triggers on: "test this web app", "E2E test", "UI test", "browser test", "验证 UI", "测试页面", "写个测试", "mock API", "端到端测试", "自动化测试", "帮我测一下", "test the login", "checkout flow", "registration flow", "dashboard test", or any mention of localhost testing, form validation, flow verification, or E2E debugging.
Generate and maintain business knowledge documents from codebase. Supports ingest (create/update docs) and lint (detect inconsistency and fix) operations. Triggers: 业务知识地图、生成业务文档、ingest domain knowledge、lint 业务知识、docify.
从原始需求素材（PRD、会议纪要、设计文档等）提炼结构化开发需求文档。通过三层模板（意图层/边界层/关联层）和五条逻辑校验规则，生成逻辑自洽的需求文档，暴露缺失和矛盾。触发词："提炼需求"、"结构化需求"、"需求文档"、"需求整理"、"refine requirement"。适用于 AI Coding Agent 开发前的需求整理，也适用于人类开发者审查和完善需求。
E2E testing methodology for local web applications — 21 principles across 5 domains (testing philosophy, data & mock, UI interaction, diagnosis & debugging, reporting). Pure methodology: no tool commands, no project conventions — teaches the agent how to think, not what to do. Use this skill whenever the user wants to test a local web app, validate UI behavior, walk through multi-step user flows, mock API responses for testing, debug E2E test failures, capture screenshots, or understand E2E testing best practices. Triggers on: "test this web app", "E2E test", "UI test", "browser test", "验证 UI", "测试页面", "写个测试", "mock API", "端到端测试", "自动化测试", "帮我测一下", "test the login", "checkout flow", "registration flow", "dashboard test", or any mention of localhost testing, form validation, flow verification, or E2E debugging.
高质量知识写作技能，创作精炼优雅，图文并茂的 Markdown 学习文章。 **当用户想要学习、研究或理解任何主题时，始终使用此技能。** 作为经验丰富的知识写作者—— 深度研究材料，识别核心概念与关系，撰写简洁、精准、有洞见的文章。 触发语（中文）："帮我学习 X"、"我想了解 X"、"研究 X"、"学习这个"、"总结"、"解释 X"、"消化这个"、"帮我深入理解" 触发语（英文）："help me learn"、"I want to understand"、"study this"、"summarize"、"teach me about" 特点：Markdown 输出配优雅图表（Mermaid/SVG/ASCII），自适应主题结构，高层写作原则， 视觉风格预览选择，内置去 AI 味处理，支持中英文输出。