agent-patterns

Agent Patterns

When and how to use each agent type effectively

Overview

This guide provides detailed patterns for using each agent type, including decision trees, real-world examples, and composition strategies for complex workflows.

---

Exploration Agents

Scout Agent

Purpose: Fast file discovery using pattern matching.

Best For:

• Finding files by name patterns
• Locating implementations before changes
• Understanding file distribution across directories

Invocation:

Task({
    subagent_type: 'scout',
    prompt: 'Find all files related to employee validation in ServiceB',
    description: 'Find employee validation files'
});

Pattern: Targeted Search

// Specific file type in specific service
Task({
    subagent_type: 'scout',
    prompt: 'Find all *CommandHandler.cs files in ServiceA/UseCaseCommands/',
    description: 'Find ServiceA command handlers'
});

Pattern: Relationship Discovery

// Find related files across boundaries
Task({
    subagent_type: 'scout',
    prompt: 'Find all files that reference EmployeeEntity - include consumers, producers, DTOs, and tests',
    description: 'Find EmployeeEntity dependencies'
});

When to Use Scout vs Direct Tools:

Need	Use Scout	Use Direct Tools
Find by name pattern	Multiple directories	Single known directory
Find by content	Cross-service	Known file
Understand structure	Yes	No
Quick lookup	No	Yes

---

Explore Agent

Purpose: Thorough codebase exploration with context.

Best For:

• Understanding how features work
• Mapping service boundaries
• Answering architectural questions

Invocation:

Task({
    subagent_type: 'Explore',
    prompt: 'How does the leave request approval workflow work? Trace from API endpoint to final state change.',
    description: 'Explore leave approval workflow'
});