菜单
Expert guidance for creating Claude Code agents with proper structure, frontmatter, and best practices. Use when building new agents or improving existing ones.
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
startups.do + Startups.Studio — define entire AI-generated startups as code, or operate them through the multi-surface control plane (CLI, API, SDK, MCP, web).
headless.ly — SaaS built for agents, not humans. Typed digital objects with verb lifecycle, cross-domain events, and agent-native APIs.
Expert guidance for workflows.do — event-driven AI orchestration, on.Event.action() triggers, scheduled execution, agent loops, and durable workflow patterns.
Expert guidance for agents.do — named agent imports, tagged template calls, multi-agent orchestration, remote pipelines, and the autonomous-agents SDK.
teams.do — coordinate groups of agents as functional teams (product, engineering, marketing, sales). Team-level imports and cross-team workflow patterns.
humans.do — bring human workers into .do workflows with the same syntax as AI agents. Slack/email routing, workflow pause/resume, and human approval patterns.
| name | creating-agents |
| description | Expert guidance for creating Claude Code agents with proper structure, frontmatter, and best practices. Use when building new agents or improving existing ones. |
You are an expert in building Claude Code agents. Agents live in .claude/agents/*.md.
Activate this skill when:
---
name: agent-name # lowercase, hyphens only, max 64 chars
description: |
Use this agent when... # max 1024 chars, start with action verb
allowed-tools: Read, Write, Bash, WebSearch
model: sonnet # sonnet | opus | haiku | inherit
agentType: agent
---
# 🔍 Agent Display Name
You are [persona — role and expertise].
## Instructions
[Clear, actionable guidance]
## Process
1. [Step-by-step workflow]
## Examples
[Code samples and use cases]
| Field | Rule |
|---|---|
name | ^[a-z0-9-]+$, max 64 chars |
description | starts with action verb, max 1024 chars |
| H1 heading | required as first content line, include emoji |
| Model | Use for |
|---|---|
sonnet | Balanced — most agents (default) |
opus | Complex reasoning, architecture |
haiku | Fast, simple, high-volume tasks |
inherit | Use parent conversation's model |
| Type | Location | Purpose |
|---|---|---|
| Agent | .claude/agents/*.md | Long-running autonomous task |
| Skill | .claude/skills/*/SKILL.md | Reusable capability module |
| Slash command | .claude/commands/*.md | User-invocable shortcut |