fusion-skill-authoring

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Create Skill

创建Skill

When to use

适用场景

Use this skill when you need to create a new skill under

skills/

and want a valid, maintainable

SKILL.md

Typical triggers:

"Create a skill for ..."
"Scaffold
```
skills/<name>/SKILL.md
```
"
"Add a new skill with proper metadata and guardrails"
"Help me create a skill that helps create other skills"

当你需要在

skills/

目录下创建新Skill，并且想要生成合法、可维护的

SKILL.md

文件时使用本Skill。

典型触发词：

"为...创建一个Skill"
"搭建
```
skills/<name>/SKILL.md
```
脚手架"
"添加带有规范元数据和防护规则的新Skill"
"帮我创建一个用于生成其他Skill的Skill"

When not to use

不适用场景

Do not use this skill for:

Editing product/application code outside
```
skills/
```
Large refactors of existing skills unless explicitly requested
Running destructive commands or making unrelated repository changes

不要将本Skill用于以下场景：

编辑
```
skills/
```
目录外的产品/应用代码
对现有Skill进行大规模重构，除非有明确要求
运行破坏性命令或对仓库进行无关修改

Required inputs

必要输入

Collect before writing files:

Base skill name in kebab-case (
```
<skill-name>
```
, without prefix)
Prefix choice (ask explicitly; suggest
```
custom-
```
by default unless the repository has its own convention)
Final skill name (
```
custom-<skill-name>
```
unless the user chooses a different prefix or none)

Target path (

skills/<final-skill-name>/

skills/.experimental/<final-skill-name>/

, or

skills/.curated/<final-skill-name>/

)

Initial semantic version for frontmatter metadata (
```
MAJOR.MINOR.PATCH
```
, default
```
0.0.0
```
for skills created in this repository)
License for frontmatter (
```
MIT
```
by default, or repository-specific choice)
One-sentence purpose for frontmatter
```
description
```
Expected output (files, commands, summary)
Safety boundaries

Validate metadata constraints:

```
name
```
: <= 64 chars, lowercase letters/numbers/hyphens only, no XML tags, and no platform-reserved words
```
description
```
: non-empty, <= 1024 chars, no XML tags, includes both what it does and when to use it
```
metadata.version
```
: semantic version string (
```
MAJOR.MINOR.PATCH
```
) in quoted YAML format
```
metadata
```
: primarily string key/value map; arrays allowed for explicit relationship fields, avoid nested objects
- ```
metadata.role
```
  :
```
"orchestrator"
```
  or
```
"subordinate"
```
  (subordinates cannot run without their orchestrator)
- ```
metadata.orchestrator
```
  : required orchestrator skill name (subordinates only)
- ```
metadata.skills
```
  : list of subordinate skill names (orchestrators only)
- ```
metadata.tags
```
  : optional list of lowercase kebab-case strings for discoverability
- ```
metadata.mcp
```
  : optional map for MCP server needs (
```
required
```
  and
```
suggested
```
  lists)
```
license
```
and
```
compatibility
```
: optional top-level frontmatter fields (not inside
```
metadata
```
)

If required inputs are missing, ask concise targeted questions first. Use

assets/follow-up-questions.md

as the default question bank.

在写入文件前请收集以下信息：

短横线命名格式的基础Skill名称（
```
<skill-name>
```
，不带前缀）
前缀选择（明确询问用户；除非仓库有自己的规范，否则默认建议使用
```
custom-
```
）
最终Skill名称（除非用户选择其他前缀或不使用前缀，否则格式为
```
custom-<skill-name>
```
）

目标路径（

skills/<final-skill-name>/

、

skills/.experimental/<final-skill-name>/

或

skills/.curated/<final-skill-name>/

）

前置元数据的初始语义化版本号（
```
MAJOR.MINOR.PATCH
```
，本仓库创建的Skill默认值为
```
0.0.0
```
）
前置元数据的许可证（默认
```
MIT
```
，或遵循仓库特定选择）
前置元数据
```
description
```
字段的一句话用途说明
预期输出（文件、命令、摘要）
安全边界

元数据约束校验：

```
name
```
：长度不超过64字符，仅允许小写字母/数字/短横线，无XML标签，无平台保留字
```
description
```
：非空，长度不超过1024字符，无XML标签，需同时包含功能说明和适用场景
```
metadata.version
```
：语义化版本字符串（
```
MAJOR.MINOR.PATCH
```
），使用YAML引号包裹格式
```
metadata
```
：主要为字符串键值对；明确的关系字段允许使用数组，避免嵌套对象
- ```
metadata.role
```
  ：
```
"orchestrator"
```
  或
```
"subordinate"
```
  （从属Skill无法脱离对应的编排Skill独立运行）
- ```
metadata.orchestrator
```
  ：必填的编排Skill名称（仅从属Skill需要）
- ```
metadata.skills
```
  ：从属Skill名称列表（仅编排Skill需要）
- ```
metadata.tags
```
  ：可选的短横线命名小写字符串列表，用于提升可发现性
- ```
metadata.mcp
```
  ：可选的MCP服务需求映射（包含
```
required
```
  和
```
suggested
```
  列表）
```
license
```
和
```
compatibility
```
：可选的顶层前置元数据字段（不放在
```
metadata
```
内部）

如果缺失必要输入，请先询问简洁的针对性问题。默认使用

assets/follow-up-questions.md

作为问题库。

Instructions

使用说明

Check whether an existing skill already covers the request:
- Run
```
npx -y skills add . --list
```
- If an existing skill matches, recommend using/updating that skill instead of creating a duplicate
- If a skill almost matches, open/recommend a repository issue to request tweaks to that existing skill instead of creating a new custom skill
- Prioritize reuse of repository skills to avoid proliferation of one-off custom skills
Confirm scope, base skill name, prefix choice, and target path.
Derive
```
<final-skill-name>
```
from prefix choice.
Create
```
<target>/<final-skill-name>/SKILL.md
```
.
Create resource directories as needed:
- ```
references/
```
  for longer guidance and detailed docs
- ```
assets/
```
  for templates/checklists/static resources
- ```
scripts/
```
  only when deterministic automation is required
Write frontmatter (
```
name
```
,
```
description
```
) that satisfies constraints.
Add the core sections:
- When to use
- When not to use
- Required inputs
- Instructions
- Expected output
- Safety & constraints
Keep
```
SKILL.md
```
concise; move long guidance/examples to
```
references/
```
.
Use
```
assets/
```
for templates, sample files, and static resources used by the skill.
Add
```
scripts/
```
only when deterministic automation is required.
When inputs are missing, ask from
```
assets/follow-up-questions.md
```
and proceed once answered.

检查是否已有现有Skill覆盖需求：
- 运行
```
npx -y skills add . --list
```
- 如果已有匹配的Skill，建议使用/更新该Skill，不要重复创建
- 如果有接近匹配的Skill，建议提交仓库Issue申请调整现有Skill，不要创建新的自定义Skill
- 优先复用仓库现有Skill，避免一次性自定义Skill泛滥
确认范围、基础Skill名称、前缀选择和目标路径。
根据前缀选择生成
```
<final-skill-name>
```
。
创建
```
<target>/<final-skill-name>/SKILL.md
```
文件。
按需创建资源目录：
- ```
references/
```
  ：存放更详细的指导和文档
- ```
assets/
```
  ：存放模板/检查清单/静态资源
- ```
scripts/
```
  ：仅在需要确定性自动化时创建
编写符合约束的前置元数据（
```
name
```
、
```
description
```
）。
添加核心章节：
- 适用场景
- 不适用场景
- 必要输入
- 使用说明
- 预期输出
- 安全与约束
保持
```
SKILL.md
```
简洁；将长指导/示例移动到
```
references/
```
目录。
将Skill使用的模板、示例文件、静态资源放在
```
assets/
```
目录。
仅在需要确定性自动化时添加
```
scripts/
```
目录。
缺失输入时，参考
```
assets/follow-up-questions.md
```
询问用户，获得答复后继续执行。

Expected output

预期输出

Return:

Created/updated file paths
Validation command(s) run
Pass/fail result with interpretation
Any required follow-up actions
Any unresolved follow-up questions (if user input was still missing)
Whether an existing skill was reused, and if an issue was created/recommended for an almost-match

For a newly scaffolded skill, the default structure should be:

text

skills/<final-skill-name>/
├── SKILL.md
├── references/
└── assets/

Use this baseline for generated

SKILL.md

files:

markdown

---
name: <final-skill-name>
description: <what it does + when to use it>
license: MIT
metadata:
   version: "<initial-version>"
---

返回以下内容：

已创建/更新的文件路径
已运行的验证命令
包含解读的通过/失败结果
所有需要的后续操作
所有未解决的后续问题（如果仍缺失用户输入）
是否复用了现有Skill，以及是否为接近匹配的Skill创建/建议提交Issue

新搭建的Skill默认结构如下：

text

skills/<final-skill-name>/
├── SKILL.md
├── references/
└── assets/

生成

SKILL.md

文件的基线模板如下：

markdown

---
name: <final-skill-name>
description: <what it does + when to use it>
license: MIT
metadata:
   version: "<initial-version>"
---

<Skill Title>

When to use

When not to use

Required inputs

Instructions

Expected output

Safety & constraints

undefined

undefined

Validation

验证

Run from repo root:

bash

npx -y skills add . --list

Pass criteria:

Exit code is
```
0
```
New skill appears discoverable/valid
No metadata/path errors

If validation fails:

Fix frontmatter (
```
name
```
,
```
description
```
)
Verify path and
```
SKILL.md
```
existence
Re-run and report final status

在仓库根目录运行：

bash

npx -y skills add . --list

通过标准：

退出码为
```
0
```
新Skill可被发现/验证通过
无元数据/路径错误

如果验证失败：

修复前置元数据（
```
name
```
、
```
description
```
）
验证路径和
```
SKILL.md
```
文件是否存在
重新运行命令并报告最终状态

Ready-to-merge checklist

合并就绪检查清单

Use

assets/ready-to-merge-checklist.md

as a lightweight PR gate.

使用

assets/ready-to-merge-checklist.md

作为轻量级PR准入条件。

Safety & constraints

安全与约束

Never:

Request or expose secrets/credentials
Run destructive commands without explicit user confirmation
Invent validation results
Modify unrelated files outside requested scope
Add hidden network-fetch or unsafe script guidance without explicit request

禁止以下行为：

请求或泄露密钥/凭证
未经用户明确确认运行破坏性命令
伪造验证结果
修改请求范围外的无关文件
未经明确要求添加隐藏网络请求或不安全脚本指引