Understanding ClaudeBoyz Agent Hierarchy

ClaudeBoyz includes 11 specialized agents that work together like a development team. Here's how they coordinate.

The Agent Hierarchy

Planner (Opus)
    └── Orchestrator (Sonnet)
            ├── Builder (Sonnet)
            ├── Debugger (Sonnet)
            ├── Marketer (Sonnet)
            ├── Writer (Sonnet)
            └── Mobile (Sonnet)
                    └── Reviewer (Haiku)

Why Different Models?

Each model has different strengths and costs:

• **Opus** - Best reasoning, highest cost. Used for security-critical decisions.
• **Sonnet** - Great balance of capability and cost. Used for most implementation.
• **Haiku** - Fast and cheap. Used for validation and quick checks.

Agent Responsibilities

Planner (Opus)

• Strategic planning and architecture
• Security-critical tasks (auth, payments, RLS)
• Complex decision-making
• Breaking down large projects

Orchestrator (Sonnet)

• Coordinating between agents
• Task delegation
• Workflow management
• Progress tracking

Builder (Sonnet)

• Feature implementation
• Code writing
• Infrastructure setup
• Integration work

Debugger (Sonnet)

• Finding and fixing bugs
• Performance issues
• Error analysis
• Troubleshooting

Marketer (Sonnet)

• Landing page copy
• Value propositions
• Competitor analysis
• Positioning

Writer (Sonnet)

• Documentation
• Blog posts
• README files
• Technical writing

Mobile (Sonnet)

• React Native development
• Expo configuration
• Firebase integration
• App Store preparation

Reviewer (Haiku)

• Code quality checks
• Build validation
• Quick verification
• Linting and formatting

Security Routing

Security-critical tasks automatically route to Planner (Opus):

• `auth-supabase` - Authentication implementation
• `auth-firebase` - Mobile authentication
• `database-rls` - Row Level Security policies
• `webhook-handler` - Payment webhook security
• `firestore-schema` - Firestore security rules

Model Selection

The Alpha agent (/claudeboyz:alpha) automatically selects models:

• Analyze task complexity
• Check for security implications
• Route to appropriate agent
• Validate with Reviewer

Cost Optimization

By using the right model for each task:

• Strategic decisions → Opus (worth the cost)
• Implementation → Sonnet (balanced)
• Validation → Haiku (fast and cheap)

This approach reduces costs while maintaining quality for critical decisions.

Using Agents Directly

You can call agents directly:

# In Claude Code with ClaudeBoyz
Use the claudeboyz-planner agent to plan this feature
Use the claudeboyz-builder agent to implement
Use the claudeboyz-reviewer agent to validate

Or use slash commands that orchestrate automatically:

/claudeboyz:launch   # Orchestrates multiple agents
/claudeboyz:alpha    # Autonomous agent selection