{
  "markdown": "# willhong-skills\n\nSkills shared by Will Hong for Claude Code and other AI agents.\n\n## Installation\n\n```bash\nnpx skills add Howell5/willhong-skills\n```\n\n## Available Skills\n\n| Skill | Description |\n|-------|-------------|\n| [graphify](#graphify) | Code navigation layer for AI agents — AST extraction from 12 languages, queryable knowledge graph |\n| [vitepress-tutorial](#vitepress-tutorial) | Generate VitePress documentation sites for source code learning |\n| [social-polish](#social-polish) | 从核心观点生成优质社媒长文，运用 4D 框架分析，去 AI 味，6 维度评审 |\n| [frontend-i18n-orchestrator](#frontend-i18n-orchestrator) | Frontend 项目 i18n 全链路编排：自动探测、最少提问、分阶段改造与质量门禁 |\n| [subtraction-audit](#subtraction-audit) | 代码减法审计与反腐化工程：证据驱动找死代码/重复实现/死依赖，风险分级执行；AI 重度仓库防膨胀门禁 |\n\n---\n\n### graphify\n\nCode navigation layer for AI coding agents. Extracts AST structure from 12 programming languages via tree-sitter WASM, builds a queryable knowledge graph, and keeps it in sync as code changes.\n\n```bash\n/graphify build         # index the current project\n/graphify query <name>  # find symbols by name\n/graphify update <file> # re-index after editing\n```\n\n**Features:**\n\n- 12 languages: Python, JS/TS, Go, Rust, Java, C/C++, Ruby, C#, Kotlin, Scala, PHP\n- Deterministic AST extraction via tree-sitter WASM — no LLM needed for structure\n- Incremental updates — only re-extracts changed files\n- Call graph, inheritance, import relationships\n- Optional semantic labeling (agent-driven, no API key)\n\n**Setup:** `npm i -g graphify-ts` (requires [Bun](https://bun.sh) runtime)\n\n**Source:** [github.com/Howell5/graphify-ts](https://github.com/Howell5/graphify-ts)\n\n---\n\n### vitepress-tutorial\n\nGenerate VitePress documentation sites for source code learning and analysis. Unlike user documentation that explains \"how to use\", these tutorials explain \"how it's implemented\".\n\n```bash\n/vitepress-tutorial [source-path] [output-path]\n```\n\n**Examples:**\n\n```bash\n/vitepress-tutorial ./apps/runner ./tutorials/runner-guide\n/vitepress-tutorial ./pkg/sandbox\n```\n\n**Features:**\n\n- Source References: Auto-generate `Source: path/to/file.go:123` annotations\n- Mermaid Diagrams: Architecture, sequence, and flow diagrams\n- Code Highlighting: Go, TypeScript, Python with line highlighting\n- Chinese-first: Content in Chinese, code comments in English\n- Standalone Deploy: Ready for Vercel, Netlify, or GitHub Pages\n\n**Workflow:**\n\n1. **Analysis** - Explore source directory, identify components and architecture\n2. **Planning** - Generate tutorial outline and chapter structure\n3. **Generation** - Create VitePress project with tutorial content\n\n**Output Structure:**\n\n```\n{output-path}/\n├── package.json\n├── docs/\n│   ├── .vitepress/\n│   │   └── config.ts\n│   ├── index.md\n│   ├── introduction/\n│   │   ├── overview.md\n│   │   └── architecture.md\n│   └── {modules}/\n│       ├── index.md\n│       └── {topics}.md\n└── README.md\n```\n\n---\n\n### social-polish\n\n从核心观点生成可直接发布的优质社媒长文。通过 4D 框架深度分析、去 AI 味处理、6 维度评审，确保输出的内容观点鲜明、情感共鸣、语言自然。\n\n```bash\n/social-polish [你的核心观点或选题]\n```\n\n**示例：**\n\n```bash\n/social-polish 为什么大多数人学不会编程\n/social-polish 远程工作三年后我发现的真相\n/social-polish 年轻人为什么越来越不想结婚\n```\n\n**特性：**\n\n- 4D 框架分析：哲学、心理学、传播学、社会学四维度拆解选题\n- 去 AI 味处理：删除空洞过渡词、模板化开头、说教式结尾\n- 6 维度评审：观点独特性、情感共鸣度、结构清晰度、语言自然度、信息密度、传播潜力\n- 自动迭代：低于 54/60 分自动修订，最多 3 轮\n\n**适用平台：** X 长推、小红书图文、微信公众号、即刻等\n\n---\n\n### frontend-i18n-orchestrator\n\n面向 React/Vue/Next/Nuxt/Vite 的项目级 i18n 改造 skill。默认先自动扫描项目，再只询问无法自动推断的业务决策，最后分阶段完成迁移与验证。\n\n```bash\n/frontend-i18n-orchestrator\n```\n\n**特性：**\n\n- Auto-first：先探测框架、现有 i18n 状态、文案热点，再提问\n- Ask-on-ambiguity：仅在业务策略不明确时提问（目标语言、URL 策略、fallback 等）\n- 分阶段执行：基建 -> 文案迁移 -> 本地化格式 -> 路由/SEO -> CI 治理\n- 质量门禁：缺 key、孤儿 key、硬编码回归检查\n- Claude Code 与 Codex 双端兼容\n\n---\n\n### subtraction-audit\n\n代码减法审计与反腐化工程。核心信念：代码最终会腐化，人写机器写都一样；防腐化唯一可靠的方式是把约束变成机器门禁、把为什么变成文档、把删除变成一等工作流。方法学源自 DeepSeek Harness。\n\n```bash\n/subtraction-audit            # 对当前仓库执行完整减法审计\n\"帮我清理死代码\"              # 或直接说意图，skill 自动加载\n```\n\n**工作流（Phase 0→6）：**\n\n1. **先懂架构**：路由挂载方式、同名活文件、部署清单 vs 实际、env 旋钮、DB 队列——声称任何东西\"死\"之前先答这些\n2. **基线测量**：knip（未使用导出/类型/文件/依赖/重复/二进制）+ 仓库卫生 + CI/钩子盘点\n3. **分域并行审计**：api / web / worker / shared / 脚本，子代理各查一类，证据先行\n4. **高危复核**：动态引用、框架自动发现误报、同名陷阱、测试是否钉死它不存在\n5. **风险分级**：🟢 证据闭环可删 / 🟡 删前确认 / 🔴 产品决策\n6. **分批执行**：每批配 DECISIONS.md 决策记录，删完 tsc + lint + knip 基线不升\n\n**关键教训（内嵌）：**\n\n- 部署清单里没有 ≠ 不需要——可能是配置漂移，甚至生产停摆（worker 教训）\n- 只删明显死符号会漏掉文件级重复生命周期最重的地方（双份 withRetry、双组件树）\n- 零 CI 是腐化积累的根因：门禁 > 散文\n\n## License\n\nMIT\n",
  "bytes": 4393,
  "sha": "fbb6c1a8a02d98db55837a716cd2812cd8304929d044055aaa7e7280ba91fdf8",
  "repo_slug": "howell5/willhong-skills",
  "fonte": "repo",
  "truncated": false,
  "api": "https://agentalog.com/api/listings/skl_howell5_willhong_skills_graphify_848347e3/readme"
}