Agents

Agents are specialized AI assistants with deep knowledge in specific domains. They act as expert team members, each with their own skills, tools, and workflows tailored to particular development tasks.

What Are Agents?

Agents are markdown files (.md) stored in .claude/agents/ that define:

Expertise area - Domain knowledge and specialization
Tools - Which Claude Code tools the agent can use
Model - Which Claude model to use (sonnet, opus, haiku)
Instructions - How the agent approaches tasks
Communication patterns - How the agent interacts with users and other agents

Agents are essentially specialized prompts that Claude Code loads when invoked. They provide consistent, expert-level assistance for specific tasks.

Agent Structure

Agents use YAML frontmatter followed by markdown instructions:

---
name: frontend-developer
description: Senior frontend developer specializing in React, Vue, and Angular
tools: Read, Write, Edit, Bash, Glob, Grep
model: sonnet
---

You are a senior frontend developer with expertise in modern web applications...

## Communication Protocol

Always begin by requesting project context from the context-manager...

## Execution Flow

1. Context Discovery
2. Development Execution  
3. Handoff and Documentation

Frontmatter Fields

Field	Required	Description
`name`	Yes	Agent identifier (kebab-case)
`description`	Yes	Clear, specific description of agent’s purpose
`tools`	Yes	Comma-separated list of allowed tools
`model`	Yes	`sonnet`, `opus`, or `haiku`

Agent Categories

Agents are organized by domain expertise:

Development Team
Security
Database
AI & Data

Core development specialists:

frontend-developer - React, Vue, Angular expert
backend-developer - API and server-side development
fullstack-developer - End-to-end application development
mobile-developer - iOS and Android applications
devops-engineer - Infrastructure and deployment

npx claude-code-templates@latest --agent development-team/frontend-developer

Security and compliance specialists:

api-security-audit - API security vulnerability assessment
security-auditor - Comprehensive security reviews
penetration-tester - Security testing and exploitation
compliance-validator - GDPR, HIPAA, SOC2 compliance

npx claude-code-templates@latest --agent security/api-security-audit

Database specialists:

postgresql-optimizer - PostgreSQL performance tuning
mysql-expert - MySQL schema design and optimization
mongodb-specialist - NoSQL database management
redis-architect - Caching and real-time features

npx claude-code-templates@latest --agent database/postgresql-optimizer

AI and data science specialists:

ml-engineer - Machine learning model development
data-scientist - Data analysis and visualization
nlp-specialist - Natural language processing
computer-vision - Image and video processing

npx claude-code-templates@latest --agent data-ai/ml-engineer

Using Agents

Installation

Install agents using the CLI:

# Single agent
npx claude-code-templates@latest --agent frontend-developer

# Multiple agents
npx claude-code-templates@latest \
  --agent frontend-developer \
  --agent backend-developer \
  --agent devops-engineer

# With category prefix
npx claude-code-templates@latest --agent development-team/frontend-developer

Invocation

Once installed, agents are available in Claude Code:

@frontend-developer Build a responsive product catalog with filtering

Or reference them in conversations:

Use the frontend-developer agent to create the React components

Agents appear in Claude Code’s agent picker automatically after installation.

Real-World Examples

Example 1: Frontend Developer Agent

File: .claude/agents/frontend-developer.md

---
name: frontend-developer
description: Use when building complete frontend applications across React, Vue, and Angular frameworks
tools: Read, Write, Edit, Bash, Glob, Grep
model: sonnet
---

You are a senior frontend developer specializing in modern web applications with deep expertise in React 18+, Vue 3+, and Angular 15+.

## Communication Protocol

### Required Initial Step: Project Context Gathering

Always begin by requesting project context from the context-manager:

```json
{
  "requesting_agent": "frontend-developer",
  "request_type": "get_project_context",
  "payload": {
    "query": "Frontend development context needed: current UI architecture, component ecosystem, design language"
  }
}

Execution Flow

Context Discovery - Map existing frontend landscape
Development Execution - Build components with TypeScript
Handoff and Documentation - Deliver with tests and docs


**Usage**:

@frontend-developer Create a dashboard with real-time data visualization using React and TypeScript

### Example 2: API Security Audit Agent

**File**: `.claude/agents/api-security-audit.md`

```markdown
---
name: api-security-audit
description: API security audit specialist. Use for REST API security audits, authentication vulnerabilities, authorization flaws
tools: Read, Write, Edit, Bash
model: sonnet
---

You are an API Security Audit specialist focusing on identifying, analyzing, and resolving security vulnerabilities in REST APIs.

## Core Expertise

- **Authentication Security**: JWT vulnerabilities, token management
- **Authorization Flaws**: RBAC issues, privilege escalation
- **Injection Attacks**: SQL injection, NoSQL injection prevention
- **Data Protection**: Sensitive data exposure, encryption
- **API Security Standards**: OWASP API Top 10
- **Compliance**: GDPR, HIPAA, PCI DSS requirements

## Security Audit Checklist

### Authentication & Authorization

```javascript
// Secure JWT implementation
const jwt = require('jsonwebtoken');

class AuthService {
  generateToken(user) {
    return jwt.sign(
      { userId: user.id, role: user.role },
      process.env.JWT_SECRET,
      { expiresIn: '15m', issuer: 'your-api' }
    );
  }
}


**Usage**:

@api-security-audit Review the authentication endpoints in src/api/auth.js

## Agent Communication Patterns

Agents can communicate with each other through structured protocols:

### Context Manager Integration

Most agents start by querying the `context-manager` for project information:

```json
{
  "requesting_agent": "agent-name",
  "request_type": "get_project_context",
  "payload": {
    "query": "Specific context needed"
  }
}

This prevents redundant questions and ensures agents have full context.

Agent-to-Agent Handoffs

Agents can delegate specialized tasks:

Status update from frontend-developer:
Components delivered. Requesting backend-developer to create API endpoints for data fetching.

Tool Permissions

Agents specify which tools they need:

Tool	Purpose
`Read`	Read files from the filesystem
`Write`	Create new files
`Edit`	Modify existing files
`Bash`	Execute shell commands
`Glob`	Find files by pattern
`Grep`	Search file contents

Only grant agents the minimum tools they need. A documentation agent doesn’t need Bash access.

Model Selection

Choose the right model for each agent:

Sonnet (Default)
Opus
Haiku

Best for: Most development tasks

Balanced performance and cost
200K context window
Strong coding capabilities
Good for agents that need to read/write code

model: sonnet

Best for: Complex analysis and architecture

Highest reasoning capability
Large context handling
Best for security audits, architecture reviews
Higher cost per request

model: opus

Best for: Simple, fast tasks

Fastest response times
Lower cost
Good for linting, formatting, simple edits
Limited context window

model: haiku

Creating Custom Agents

You can create your own agents:

Create Agent File

Create .claude/agents/my-agent.md

Add Frontmatter

Define name, description, tools, and model

Write Instructions

Provide clear, specific instructions for the agent’s behavior

Test

Invoke the agent in Claude Code: @my-agent Task description

For detailed guidance on creating agents, see Creating Components.

Agent Best Practices

1. Clear Scope Definition

Define exactly when to use the agent:

## When to Use This Agent

Use this agent for:
- Comprehensive API security audits
- Authentication and authorization reviews
- Vulnerability assessments

Do NOT use for:
- General code review (use code-reviewer)
- Frontend security (use frontend-security-specialist)

2. Context Gathering

Always gather context before starting work:

Query context-manager for project information
Check existing patterns and conventions
Validate assumptions with targeted questions

3. Structured Communication

Provide clear status updates:

{
  "agent": "agent-name",
  "update_type": "progress",
  "current_task": "Task description",
  "completed_items": ["Item 1", "Item 2"],
  "next_steps": ["Next task"]
}

4. Documentation

Document all deliverables:

File locations and purposes
Integration points
Configuration changes
Next steps

Global Agents

Create agents that work across all projects:

# Create global agent
npx claude-code-templates@latest --create-agent my-global-agent

# List global agents
npx claude-code-templates@latest --list-agents

# Remove global agent
npx claude-code-templates@latest --remove-agent my-global-agent

Global agents are stored in ~/.claude/agents/ and available everywhere.

Next Steps

Browse Agent Library

Explore 400+ available agents

Create Custom Agents

Build your own specialized agents

Commands

Learn about custom slash commands

Global Agents

Create globally accessible agents

Get Started

Core Concepts

Guides

Tools

Advanced

What Are Agents?

Agent Structure

Frontmatter Fields

Agent Categories

Using Agents

Installation

Invocation

Real-World Examples

Example 1: Frontend Developer Agent

Execution Flow

Agent-to-Agent Handoffs

Tool Permissions

Model Selection

Creating Custom Agents

Agent Best Practices

1. Clear Scope Definition

2. Context Gathering

3. Structured Communication

4. Documentation

Global Agents

Next Steps

Browse Agent Library

Create Custom Agents

Commands

Global Agents

Build docs developers (and LLMs) love

Get Started

Core Concepts

Guides

Tools

Advanced

​What Are Agents?

​Agent Structure

​Frontmatter Fields

​Agent Categories

​Using Agents

​Installation

​Invocation

​Real-World Examples

​Example 1: Frontend Developer Agent

​Execution Flow

​Agent-to-Agent Handoffs

​Tool Permissions

​Model Selection

​Creating Custom Agents

​Agent Best Practices

​1. Clear Scope Definition

​2. Context Gathering

​3. Structured Communication

​4. Documentation

​Global Agents

​Next Steps

Browse Agent Library

Create Custom Agents

Commands

Global Agents

Build docs developers (and LLMs) love

What Are Agents?

Agent Structure

Frontmatter Fields

Agent Categories

Using Agents

Installation

Invocation

Real-World Examples

Example 1: Frontend Developer Agent

Execution Flow

Agent-to-Agent Handoffs

Tool Permissions

Model Selection

Creating Custom Agents

Agent Best Practices

1. Clear Scope Definition

2. Context Gathering

3. Structured Communication

4. Documentation

Global Agents

Next Steps