agent-creator
Create Claude Code subagents. Use when user wants to create a subagent, specialized agent, or custom AI assistant for Claude Code.
When & Why to Use This Skill
The Agent Creator skill empowers developers to build and deploy specialized Claude Code subagents, streamlining complex workflows through custom-tailored AI assistants. By defining specific roles, tool permissions, and system prompts in a structured markdown format, users can automate repetitive tasks and create a modular ecosystem of AI agents optimized for code reviews, security audits, and project-specific automation.
Use Cases
- Automated Code Review: Create a dedicated subagent to analyze git diffs for code quality, security vulnerabilities, and maintainability standards after every change.
- Specialized Security Auditing: Deploy an agent with restricted tool access and specific system prompts to scan codebases for sensitive information and compliance protocols.
- Project-Specific Documentation: Build a custom assistant that understands your unique architecture to generate and update technical documentation or README files automatically.
- Bug Triaging and Root Cause Analysis: Configure a subagent to investigate reported issues by searching the codebase and identifying potential logic errors or regressions.
| name | agent-creator |
|---|---|
| description | Create Claude Code subagents. Use when user wants to create a subagent, specialized agent, or custom AI assistant for Claude Code. |
Agent Creator
Creates Claude Code subagents - specialized AI assistants defined as markdown files.
What is a Subagent?
A subagent is a markdown file with YAML frontmatter that defines a specialized AI assistant. Claude Code delegates tasks to subagents based on their description.
File Format
---
name: agent-name
description: When to use this agent. Include trigger words.
tools: Read, Grep, Glob
model: sonnet
---
System prompt goes here. Define the agent's role, process, and output format.
Configuration Fields
| Field | Required | Description |
|---|---|---|
name |
Yes | Lowercase letters and hyphens only |
description |
Yes | When Claude should use this agent |
tools |
No | Comma-separated list. Omit to inherit all tools |
model |
No | haiku, sonnet, opus, or inherit |
Available Tools
Read, Write, Edit, Bash, Glob, Grep, WebFetch, WebSearch, plus any MCP tools.
File Locations
| Location | Scope |
|---|---|
.claude/agents/ |
Current project only |
~/.claude/agents/ |
All projects for this user |
Creating a Subagent
Ask user for:
- Agent name
- Purpose and when to trigger
- Required tools (minimal set)
- Model choice
Create the markdown file with:
- Clear, specific description with trigger words
- Focused system prompt
- Minimal tool permissions
Save to the location user specifies
Best Practices
- Focused purpose: One clear responsibility per agent
- Specific triggers: Include action words in description ("Use when...", "Proactively...")
- Minimal tools: Only grant necessary tools
- Clear output format: Define expected response structure in prompt
Example: Code Reviewer
---
name: code-reviewer
description: Reviews code for quality and security. Use after code changes.
tools: Read, Grep, Glob, Bash
model: sonnet
---
You are a code reviewer. When invoked:
1. Run git diff to see changes
2. Review for quality, security, maintainability
3. Provide feedback by priority: Critical > Warnings > Suggestions