Skill seekers doc to skill
Skill findscripter/everything-skills/04-ai/skill-seekers-doc-to-skill
当需要把文档网站、GitHub 仓库、PDF 或视频批量转成可用的 AI 技能(SKILL.md + references)时使用;用 skill-seekers 完成抓取→增强→打包→上传/安装到 Claude 或各 Agent 全流程;不适用于手写单条精炼技能、或源缺乏结构化文档时。触发词:文档转技能、爬文档建技能、skill-seekers、批量生成技能From its SKILL.md
npx -y skills add findscripter/everything-skills --skill skill-seekers-doc-to-skillAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
2 things to look at
- no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
- 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
What its file declares
Copied from the file, not written here
The file declares its own license as MIT. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.
SKILL.md
6.4 KB, ~1.8k tokens by cl100k_base, as published. Nobody here has run it
适配自第三方工具 Skill Seekers(CLI 名
skill-seekers)。源 SKILL.md 仅为占位骨架,实质能力来自上游仓库,本条已补全真实命令与流程。
何时使用
适用:
- 某框架/SDK/产品有成体系的在线文档、GitHub 仓库、PDF 手册或教学视频,你想一次性把它变成可被 Agent 检索的技能(SKILL.md + 分类 references)。
- 需要把同一份知识打包上传到 Claude,或安装进 Cursor/Windsurf 等编码 Agent。
- 想用预设配置或 AI 扫描,对一个代码仓库自动识别框架并生成技能。
不该用(负边界):
- 你只想手写一条短小、精炼、强观点的技能——直接写,本工具产出的是文档化大块知识,不是精炼操作手册。
- 源没有结构化文档(散落聊天记录、零散截图),抓取价值低。
- 需要逐字翻译/合规改写版权文档——本工具只做结构化抽取,不解决授权问题。
步骤
- 安装(Python 3.10+):
pip install skill-seekers # 基础 pip install skill-seekers[all] # 全功能(含视频/多 LLM) skill-seekers-setup # 可选:交互式初始化 skill-seekers doctor # 自检环境 - 抓取生成(按源类型四选一/多):
skill-seekers create https://docs.react.dev/ --name react # 文档网站 skill-seekers create facebook/react --include-issues # GitHub 仓库 skill-seekers create --pdf manual.pdf --name myskill --ocr # PDF(扫描件加 --ocr) skill-seekers create --video-url <youtube> --name tut # 视频 skill-seekers scan ./my-app --out ./configs/ --enhance # AI 扫描仓库识别框架 - 增强(可选但推荐,补示例/工作流):
skill-seekers enhance output/react/ skill-seekers enhance output/react/ --workflow security-focus - 打包为目标平台(默认 Claude):
skill-seekers package output/react/ --target claude # 其他:--target gemini|openai|langchain|llama-index|markdown skill-seekers package output/react/ --chunk-for-rag --chunk-tokens 512 # RAG 切块 - 上传 / 安装:
export ANTHROPIC_API_KEY=sk-ant-... skill-seekers upload output/react.zip # API 自动上传 # 或手动:访问 https://claude.ai/skills 选择 zip skill-seekers install-agent output/react/ --agent cursor # 装进编码 Agent
指令
- 一键全流程(取配置→抓取→增强→打包→上传):
skill-seekers install --config react skill-seekers install --config django --no-upload skill-seekers install --config react --dry-run # 先空跑看计划 - 生命周期/续跑(大型抓取必备):
skill-seekers create <src> --resume # 断点续抓 skill-seekers create <src> --fresh # 忽略缓存重抓 skill-seekers create <src> --skip-scrape # 复用已抓数据只重建 skill-seekers create <src> --preset quick|comprehensive skill-seekers resume --list # 查看可续跑会话 - 质量/维护:
skill-seekers quality output/react/、skill-seekers update output/react/、skill-seekers enhance-status output/react/ --watch。 - 提速:
--async --workers 8(约快 2–3 倍)。
示例
把 Django 文档做成技能并装进 Windsurf,不上传 Claude:
skill-seekers create https://docs.django.com/ --name django --async --workers 8
skill-seekers enhance output/django/
skill-seekers package output/django/ --target claude
skill-seekers install-agent output/django/ --agent windsurf
产出结构:
output/
├── django_data/ # 原始抓取(pages/*.json + summary.json)
└── django/
├── SKILL.md # 已增强,含示例
├── references/ # 分类文档(index.md, getting_started.md, ...)
├── scripts/ # 留空,供你补充
└── assets/ # 留空
打包后:output/django-claude.zip。
注意事项
- 凭证:上传 Claude 需
ANTHROPIC_API_KEY;抓 GitHub 强烈建议设GITHUB_TOKEN——匿名仅 60 次/小时,认证后 5000 次/小时。其他平台对应GOOGLE_API_KEY/OPENAI_API_KEY;自建端点用ANTHROPIC_BASE_URL。 - SPA 文档走三层发现(sitemap → llms.txt → 无头渲染),动态站点也能抓但更慢。
- 视频需
pip install skill-seekers[video];扫描版 PDF 必须加--ocr。 - 配置/工作流/设置分别在
./configs/、~/.config/skill-seekers/workflows/、~/.config/skill-seekers/config.json。 - 生成的 SKILL.md 是「文档化」而非「精炼操作化」:自动产物偏全、偏长,入库前建议人工删冗、补强观点边界,再纳入本大典。
- 大型站点先
--dry-run估算规模,必要时--preset quick试跑。
互见
- 手写精炼技能、定义本大典 SCHEMA:见仓库 ROADMAP 与「技能编写规范」。
- 把成品技能纳入飞书侧封装:见
lark-skill-maker。
采编自 sickn33/antigravity-awesome-skills(MIT),上游工具 Skill Seekers(yusufkaraaslan/Skill_Seekers)。
What ships with it
Read from the repository
Just SKILL.md. No reference files, no scripts.