web-cliplibrary
将网页查找结果保存到持久化剪辑库，而不是聊天记录中：每条剪辑记录原始 URL、获取日期、逐字摘录，并为任何释义加上引号，同时标注项目标签——每个项目对应一个 Markdown 文件，便于 grep 搜索和引用，即使源页面发生更改或失效后仍可使用。
插件会安装到这里；不确定时保持 web。
npx -y @deepseek-ai/dsh plugin --profile web add github:ChenneyZhuang/web-cliplibrary#9fe2ec00c97db72f951eed4987aee187300fad78Web Cliplibrary 以 web-cliplibrary 发布，当前版本为 0.1.0。Plugin Hub 会校验它的 manifest，并保存精确安装来源，便于复现安装结果。
Research that lives in chat scrollback is research lost. Clip it to disk.
住在聊天记录里的调研等于丢了。剪进磁盘。
Agents read a dozen pages per research task; the facts that mattered end up paraphrased in conversation, and three days later nobody can say which page said what — or the page changed, or died. A clip library fixes the whole chain: verbatim excerpts with the URL and fetch date, so every later sentence traces back to a frozen snapshot of the source.
agent 一次调研要读十几个页面；关键事实最后只剩对话里的转述——三天后没人说得清哪个页面说了什么，或者页面已经改版、已死。剪藏库修好整条链路：逐字摘录 + URL + 抓取日期，后面每个句子都能追溯到一个被冻结的源快照。
### <page title>
- url: https://…
- fetched: 2026-09-14
- via: search / linked from X / direct
- tags: project-y, pricing
> verbatim excerpt, exactly as served …
note (mine): paraphrase or thought — clearly separated from the quote
One file per project (clips/<project>.md), newest on top, greppable. A quote is verbatim and quoted; anything rewritten is a note (mine) line. Citing a clip in a deliverable carries the URL and fetch date with it.
每个项目一个文件（clips/<project>.md），新的在上，可 grep。引文逐字且带引号；任何改写都是 note (mine) 行。交付物里引用剪藏时带上 URL 与抓取日期。
npx skills add ChenneyZhuang/web-cliplibrary
Per-agent paths: COMPATIBILITY.md. MIT. v0.1.0.
各 agent 安装路径见 COMPATIBILITY.md。MIT 许可，v0.1.0。