用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt;运行前请先检查来源。
npx skills add https://github.com/y3owk1n/nix-system-config-v2 --skill find-skills命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地?可下载 SkillsMP 当前能够提供的文件。
Create, derive, or update the PRD — the single source of truth every other skill reads.
正在显示 SKILL.md
| name | find-skills |
| description | Search the open skills ecosystem for a skill that already does what the user is about to build. |
Discovery only. Installation in this repo is declarative — /add-skill writes the flake input and the enable list.
npx skills add -g writes into ~/.claude/skills, which nix owns and regenerates. An install done that way disappears on the next just rebuild. Use it to inspect a candidate, never to keep one.
"React performance" finds things; "make it faster" does not. If the request is vague, sharpen it before searching — one specific query beats three broad ones.
skills.sh ranks by installs, which surfaces the battle-tested options before any search does.
npx skills find <query> [--owner <owner>]
Recommend nothing on search rank alone. A skill runs with your permissions and your context.
anthropics, vercel-labs, cursor, and named practitioners carry reputation. An unknown author does not.Give the user: what it does, installs and source, what you thought of the SKILL.md, and the skills.sh link.
If they want it, dispatch /add-skill with the repo and subdir. That is what makes it survive a rebuild.
Say so and do the task directly. If it is something they do often, /add-skill also creates a local skill — and /writing-for-agents covers what goes in it.
Done when: the user has a named candidate they have accepted or rejected, and any accepted skill went through /add-skill rather than npx skills add. Checkable: grep <skill> modules/home/packages/skills.nix finds it, or the user declined.