菜单
Complete reference for Claude Code skill YAML frontmatter fields, their values, and best practices. Use when writing frontmatter for SKILL.md, configuring skill invocation, setting allowed-tools, or deciding on skill metadata fields.
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。
基于 SOC 职业分类
Build frontend interfaces with externally randomized style direction. Use when the user asks to build web components, pages, or applications with high design diversity. Produces distinctive, production-grade UI that avoids repetitive AI aesthetics.
Analyzes CLAUDE.md, AGENTS.md, and similar repository context files for effectiveness based on peer-reviewed research. Identifies instructions that hurt agent performance, redundant content, and unnecessary requirements. Use when reviewing a CLAUDE.md, auditing an AGENTS.md, optimizing context files, or when the user says "lint my CLAUDE.md", "review my AGENTS.md", "is my context file good", "optimize my context file", or "check my CLAUDE.md".
Interactive guide for creating new Claude Code skills from scratch. Use when building a new skill, creating a SKILL.md, designing skill structure, or when the user says "create a skill", "build a skill", "make a new skill", or "help me write a skill".
Common patterns and architectures for building Claude Code skills. Use when deciding how to structure a skill, choosing between skill patterns, designing multi-step workflows, or when the user asks "what kind of skill should I build" or "how should I structure this skill".
Guide for packaging Claude Code skills as distributable plugins. Use when creating a plugin, packaging skills for sharing, setting up a plugin marketplace, or when the user says "make a plugin", "package my skills", "distribute skills", or "create a marketplace".
Reviews and validates existing Claude Code skills for correctness, quality, and best practices. Use when reviewing a skill, auditing SKILL.md files, checking skill quality, or when the user says "review this skill", "validate my skill", "check my SKILL.md", or "is this skill correct".
| name | skill-creator-frontmatter |
| description | Complete reference for Claude Code skill YAML frontmatter fields, their values, and best practices. Use when writing frontmatter for SKILL.md, configuring skill invocation, setting allowed-tools, or deciding on skill metadata fields. |
The YAML frontmatter between --- markers at the top of SKILL.md controls how Claude discovers and runs your skill. The frontmatter is loaded into Claude's system prompt (Level 1 of progressive disclosure), so keep it concise.
---
name: your-skill-name
description: What it does. Use when user asks to [specific phrases].
---
Both --- delimiters must appear on their own lines. Everything between them is parsed as YAML.
Display name and slash-command identifier.
# CORRECT
name: deploy-staging
name: api-conventions
name: fix-issue
# WRONG
name: Deploy Staging # spaces and capitals
name: deploy_staging # underscores
name: DeployStaging # camelCase
name: claude-helper # reserved prefix
What the skill does and when Claude should load it. This is the most impactful field.
< >)[What it does] + [When to use it] + [Key capabilities]# GOOD — specific, with trigger phrases
description: >-
Generates API documentation from source code with consistent formatting.
Use when writing API docs, creating endpoint references, or when the user
says "document this API" or "generate docs". Covers REST and GraphQL.
# GOOD — includes negative triggers to prevent over-triggering
description: >-
Advanced statistical analysis for CSV datasets. Use for regression,
clustering, and hypothesis testing. Do NOT use for simple data viewing
or chart creation (use data-viz skill instead).
# BAD — too vague
description: Helps with projects.
# BAD — no trigger conditions
description: Creates sophisticated multi-page documentation systems.
Prevents Claude from automatically loading this skill.
falsetrue, only the user can invoke via /skill-name. Description is NOT loaded into context.Use for skills with side effects: deploy, send messages, delete resources, commit code.
disable-model-invocation: true
Controls whether the skill appears in the / menu.
truefalse, skill is hidden from the user menu. Only Claude can invoke it. Description IS loaded into context.Use for background knowledge that isn't actionable as a command.
user-invocable: false
| Configuration | User invokes | Claude invokes | Description in context |
|---|---|---|---|
| (default) | Yes | Yes | Yes |
disable-model-invocation: true | Yes | No | No |
user-invocable: false | No | Yes | Yes |
Restricts which tools Claude can use when the skill is active.
Read, Write, Edit, Bash, Grep, Glob, Agent, WebFetch, WebSearchBash(python *), Bash(npm *), Bash(gh *)# Read-only exploration
allowed-tools: Read, Grep, Glob
# Script execution only
allowed-tools: "Bash(python:*) Bash(npm:*) WebFetch"
# Full access (default — omit field entirely)
Controls execution environment.
fork (only supported value)Only use context: fork when the skill has explicit task instructions. Reference-only skills should NOT use fork.
context: fork
Which subagent type to use when context: fork is set.
Explore, Plan, general-purpose, or custom agent name from .claude/agents/general-purposecontext: forkcontext: fork
agent: Explore # Read-only codebase exploration
Hint shown during / autocomplete to indicate expected arguments.
argument-hint: "[issue-number]"
argument-hint: "[filename] [format]"
argument-hint: "[component-name] [from-framework] [to-framework]"
Override the model used when this skill is active.
model: claude-sonnet-4-6
Hooks scoped to this skill's lifecycle.
License for open-source distribution.
MIT, Apache-2.0Environment requirements.
Custom key-value pairs for any additional information.
metadata:
author: Your Name
version: 1.0.0
mcp-server: server-name
category: productivity
tags: [automation, deployment]
Forbidden in frontmatter:
< >) — frontmatter appears in Claude's system promptWhy: Frontmatter is injected into the system prompt. Malicious content could be used for prompt injection.