Create a new subagent by generating a Markdown file in .claude/agents/.

Process

1. Gather Requirements

If the user provided $ARGUMENTS, use it as the agent name. Otherwise ask:

•Name: lowercase, hyphens only (e.g., code-reviewer, test-runner)
•Purpose: What task does this agent handle?
•Language: What language should the agent respond in? (default: Korean with English technical terms)

2. Determine Configuration

Based on the purpose, decide:

Field	How to decide
`model`	`haiku` for fast search/lookup, `sonnet` for general tasks, `opus` for complex reasoning
`tools`	Only grant what's needed. Read-only tasks → `Read, Glob, Grep, WebSearch, WebFetch`. Code changes → add `Edit, Write, Bash`

3. Write the Agent File

Create .claude/agents/{name}.md with this structure:

markdown

---
name: {name}
description: "{1-2 sentence description of when Claude should delegate to this agent. Include example triggers.}"
model: {model}
---

{System prompt: role definition, methodology, output format, guidelines}

4. Description Field Guidelines

The description field is critical — Claude uses it to decide when to auto-delegate. Write it to include:

•When to use this agent (trigger conditions)
•Example user messages that should activate it (use <example> blocks)
•Format examples in the pattern used by existing agents in .claude/agents/

5. System Prompt Guidelines

The body after frontmatter is the agent's system prompt. Include:

•Role definition: Who is this agent? What expertise does it have?
•Core mission: One clear sentence on what it does
•Methodology: Step-by-step process the agent should follow
•Output format: How to structure the response
•Operational guidelines: Rules and constraints
•Language: Match the project convention (Korean with English technical terms)

6. Verify

After creating the file, read it back and confirm with the user.

Reference

Check existing agents in .claude/agents/ for style consistency before writing.