AGENT SKILL

create-skill

Interactive skill scaffolder — creates a new .claude/skills/<name>/SKILL.md with proper frontmatter

License 许可
MIT
所属仓库
Linkbreathe/Mark-I
最近更新
2026-07-04
Stars
★ 0
Forks
0

01 前言 Front Matter

name
create-skill
context
inline
description
Interactive skill scaffolder — creates a new .claude/skills/<name>/SKILL.md with proper frontmatter
when-to-use
User wants to create a new skill, or asks "how do I make a skill"
user-invocable
true

03 SKILL.md 正文

Create-Skill Scaffolder

The user wants to create a new skill. Walk through this checklist interactively:

1. Ask for the skill name

  • This becomes the directory name under .claude/skills/
  • Use kebab-case (e.g. deploy-ecs, format-code)

2. Decide if it's user-invocable

  • true → user types /skill-name in REPL
  • false → only AI auto-invokes via the skill tool

3. Decide context mode

  • inline (default): prompt injected directly into conversation. Best for simple, quick tasks.
  • fork: runs in a sub-agent with isolated context. Best if the skill needs to read files, run shell commands, or do heavy research without bloating the main conversation.

4. (Optional) Limit tools for fork mode

If fork mode, ask which tools the sub-agent should be allowed to use. Common patterns:

  • Read-only: ["read_file", "list_files", "grep_search"]
  • Full tools: omit allowed-tools entirely

5. Write the prompt template

Guide the user to describe what the skill should do. Use $ARGUMENTS for user input, ${CLAUDE_SKILL_DIR} for the skill's own directory.

6. Create the file

Use write_file to write .claude/skills/<name>/SKILL.md. No other files needed — the SKILL.md is everything.

7. Verify

Run /skills to confirm it shows up in the list. If it's user-invocable, test it right away.


Example output you should produce (fill in values from the conversation):

---
name: my-skill
description: Does XYZ
user-invocable: true
context: inline
---

# My Skill

Do XYZ using $ARGUMENTS

04 同仓库更多 Skills