Custom Agents

Extensible

Create specialized AI agents tailored to your specific needs. Extend Agentwise with domain experts, tool specialists, or workflow automators.

Quick Start: Create Your First Agent

bash
# Generate a new specialized agent
/generate-agent "blockchain-specialist"

# Agent will be created at:
# .claude/agents/blockchain-specialist.md

# Use in your next project:
/create "DeFi lending platform"
# Blockchain specialist will automatically be included!

Types of Custom Agents

Domain Specialist

Expert in specific domains like healthcare, finance, or legal

Examples:

healthcare-compliance
financial-analyst
legal-reviewer

Use case: When you need specialized domain knowledge

Language Expert

Specialized in specific programming languages or frameworks

Examples:

rust-specialist
swift-developer
kotlin-expert

Use case: For language-specific optimizations

Tool Integrator

Integrates with specific tools or platforms

Examples:

salesforce-connector
sap-integrator
jira-manager

Use case: When working with enterprise tools

Process Automator

Automates specific workflows or processes

Examples:

ci-cd-optimizer
release-manager
deployment-specialist

Use case: For workflow automation

Agent Creation Process

1

Define Agent Purpose

Clearly define what your agent will specialize in

bash
/generate-agent "blockchain-specialist"
2

Agent File Creation

System creates agent definition in .claude/agents/

bash
.claude/agents/blockchain-specialist.md
3

Configure Capabilities

Define agent's skills, tools, and MCPs

bash
capabilities:
  - Smart contract development
  - DeFi protocols
  - Web3 integration
4

Automatic Integration

Agent is automatically discovered and integrated

bash
Agent ready for use in projects

Agent Definition Structure

Example: blockchain-specialist.md

Located in .claude/agents/ folder

markdown
# Blockchain Specialist Agent

## Role
Expert in blockchain development, smart contracts, and Web3 technologies.

## Capabilities
- Smart contract development (Solidity, Rust)
- DeFi protocol implementation
- NFT marketplace creation
- Cross-chain bridge development
- Gas optimization strategies
- Security auditing

## Tools & Technologies
- Ethereum, Polygon, Arbitrum
- Hardhat, Truffle, Foundry
- Web3.js, Ethers.js
- IPFS, Arweave
- TheGraph, Chainlink

## MCP Integrations
- ethereum-mcp
- web3-tools-mcp
- defi-protocols-mcp

## Best Practices
- Always implement reentrancy guards
- Use OpenZeppelin contracts when possible
- Optimize for gas efficiency
- Implement comprehensive testing
- Follow security best practices

## Phase Approach
1. Architecture design
2. Smart contract development
3. Frontend integration
4. Testing & auditing
5. Deployment & monitoring

Real-World Agent Examples

Mobile Specialist

Expert in React Native, Flutter, and native mobile development.

bash
/generate-agent "mobile-specialist"

ML Engineer

Specializes in machine learning, AI models, and data pipelines.

bash
/generate-agent "ml-engineer"

Security Auditor

Performs security audits, penetration testing, and vulnerability assessment.

bash
/generate-agent "security-auditor"

System Integration

How Custom Agents Work with Agentwise

Automatic Discovery

Agents are automatically discovered from the .claude/agents/ folder every 5 seconds. No registration required.

typescript
// Auto-discovery in action
const agents = await discoverAgents();
// Includes all custom agents

Smart Selection

Custom agents are intelligently selected based on project requirements, just like built-in agents.

typescript
// Project: "blockchain app"
// Selected: blockchain-specialist
// Automatic inclusion

MCP Support

Custom agents can use any of the 61 integrated MCP servers or define their own custom MCPs.

typescript
// Agent can use MCPs
mcp: ['ethereum-mcp', 'web3-mcp']

Token Optimization

Custom agents benefit from the same 30-40% token optimization as built-in agents.

typescript
// Shared context & caching
// Automatic optimization

Best Practices

Agent Design

  • Keep agents focused on specific domains
  • Define clear capabilities and limitations
  • Include relevant MCP integrations
  • Document best practices for the domain

Agent Collaboration

  • Design agents to work with existing ones
  • Avoid overlapping responsibilities
  • Use shared context effectively
  • Follow phase management patterns