Agent Design

Purpose

Design effective AI agents for the CMAT (Claude Multi-Agent Template) system by selecting appropriate tools, skills, and defining clear personas that enable agents to accomplish specific software development tasks.

When to Use

• Creating new agents for workflows
• Defining agent capabilities and responsibilities
• Selecting tools and skills for agent tasks
• Designing agent collaboration patterns
• Refining existing agent definitions

Key Capabilities

1. Capability Mapping - Match requirements to agent tools and skills
2. Persona Design - Define clear agent roles and instructions
3. Tool Selection - Choose appropriate tools for agent capabilities

Approach

1. Understand the Need - What task should this agent accomplish?
2. Define Role & Responsibilities - What is the agent's purpose and scope?
3. Select Tools - What capabilities does the agent need?

- Read/Write/List: File operations - Glob/Grep: Code searching - Edit/MultiEdit: Code modifications - Bash: Running commands/tests - WebSearch/WebFetch: External research

1. Choose Skills - What domain expertise is needed?
2. Write Persona - Clear instructions, examples, best practices
3. Define Outputs - What artifacts should the agent produce?

Example

Context: Need an agent to analyze API designs

Agent Design:

---
name: API Reviewer
role: API Design Analysis & Review
description: Reviews API designs for REST compliance and best practices
tools:
  - Read       # Read API specs
  - Write      # Write review reports
  - Grep       # Search for patterns
  - WebSearch  # Research API standards
skills:
  - api-design
  - technical-writing
---

# API Reviewer Agent

You are an expert API architect reviewing API designs.

## Your Responsibilities
- Analyze API endpoint designs
- Check REST compliance
- Identify inconsistencies
- Suggest improvements

## Review Checklist
1. RESTful design (nouns, not verbs)
2. Proper HTTP methods
3. Consistent error handling
4. Clear documentation

## Output
Create a review report in markdown with:
- Summary assessment
- Specific findings
- Recommendations

Why These Choices:

• Tools: Read (specs), Write (reports), Grep (patterns), WebSearch (standards)
• Skills: api-design (domain expertise), technical-writing (clear reports)
• Persona: Focused on review checklist and report format

Best Practices

• ✅ Give agents single, clear responsibilities
• ✅ Match tools to actual capabilities needed
• ✅ Include relevant skills for domain knowledge
• ✅ Provide examples and templates in persona
• ✅ Define expected output format clearly
• ✅ Keep personas focused (200-500 lines)
• ❌ Avoid: Agents with too many unrelated responsibilities
• ❌ Avoid: Giving all tools "just in case"
• ❌ Avoid: Vague persona instructions