Build, modify, optimize, or review AI agent skills — cross-platform and ecosystem-agnostic, with a low-friction interaction model. / 跨平台、不绑定生态的 Skill 制作器:创建、优化、审查智能体技能,交互低打扰。
English | 中文说明
A meta-skill for making skills. It turns repetitive workflows and ideas into well-structured skill packages, improves existing ones, and reviews drafts — following the philosophy of minimal back-and-forth and capability-probing with graceful degradation.
- Light path (default) / 轻量路径(默认) — capture intent → write SKILL.md → quick self-test → deliver. At most 2 clarifying questions, with sensible defaults stated up front
- Full path (optional) / 完整路径(可选) — adds evaluation: real test prompts, with-skill vs. without-skill baseline comparison, benchmark reruns. Only for production-grade skills with objectively verifiable outputs
- Environment probing / 环境探测 — silently detects host capabilities (parallel subtasks, CLI model calls, browser preview, Python filesystem); anything missing degrades to a fallback instead of failing
- Low-friction interaction / 低打扰交互 — merge questions into one structured confirmation; provide defaults by skill type; research instead of asking; run first, review after
- Trigger-word engineering / 触发词工程 — the
descriptionfield is the primary trigger mechanism; a dedicated guide covers the four elements (what / when / capability / trigger phrases) and anti-undertrigger wording - Structure guide / 结构指南 — directory layout, progressive disclosure, writing style for SKILL.md bodies
- Quality checklist / 质量自检 — run before delivery; light path checks core items, full path checks all
- Safe improvement / 安全优化 — existing skills get diagnosed with evidence (P1/P2/P3), confirmed before editing, backed up before changes, re-tested after
├── SKILL.md # Main skill instructions / 主指令
└── references/
├── skill-structure-guide.md # Structure & writing style / 结构与写作风格
├── trigger-writing-guide.md # Description/trigger writing / 触发词写作
└── quality-checklist.md # Pre-delivery checklist / 交付前自检
Drop the folder into your agent's skills directory. Use when you want to "turn this workflow into a skill", "帮我做个 skill", optimize or review an existing skill — even if the word "skill" is never mentioned.
一个「做技能的技能」。把重复工作流和想法固化为结构良好的 skill 包,也能优化和审查已有 skill。设计哲学:交互低打扰、先探测能力再降级。
- 轻量路径(默认):抓意图 → 写 SKILL.md → 快速自测 → 交付。反问 ≤ 2 次,默认值先给足
- 完整路径(可选):加评测——真实测试 prompt、带/不带 skill 基线对照、benchmark 重跑。只用于要上生产、输出客观可验证的 skill
- 环境探测:静默探测宿主能力(并行子任务、CLI 调用、浏览器、Python),缺什么就降级,不报错
- 低打扰交互:问题合并成一次结构化确认;按类型给默认值;能查不问;先跑再评价
- 触发词工程:
description是首要触发机制,专门指南覆盖四要素与反欠触发写法 - 结构指南:目录结构、渐进披露、SKILL.md 写作风格
- 质量自检:交付前必跑;轻量查核心项,完整查全部
- 安全优化:证据说话(P1/P2/P3 分级)、改前确认、改前备份、改后复测
放入智能体 skills 目录。想「把重复流程做成 skill」「帮我做个 skill」「优化/审查一个 skill」时使用——即使没说出 skill 这个词。