C-Breakout: Distributed Task Execution for Claude Code

Intelligent project orchestration with specialized agents. Reduces token overhead by 43-50% through agent specialization.

What is C-Breakout?

C-Breakout is an advanced orchestration system for Claude Code that breaks down complex projects into specialized agents, each with minimal context overhead. Instead of spawning full-context agents (19K tokens each), it uses an intelligent orchestrator to assign tasks to specialized agent types (8-12K tokens each).

Key Features

• 🎯 Intelligent Orchestration: Analyzes projects and creates optimal execution plans
• 💰 43-50% Cost Reduction: Specialized agents use fewer tokens
• 🚀 Parallel Execution: Wave-based execution respects dependencies
• 🔒 Better Security: Agents only get tools they need (principle of least privilege)
• 📊 Progress Monitoring: Real-time tracking of agent execution
• 🔄 Automatic Retry: Handles failures with configurable retry logic

Installation

npm install -g @claude-code/c-breakout

The skill will be automatically installed to ~/.claude/commands/c-breakout.md

Quick Start

In Claude Code

/c-breakout orchestrate "Implement user authentication system"

Natural Conversation

Orchestrate implementing a REST API for products

Claude will recognize your intent and invoke the orchestrator automatically.

How It Works

Architecture

┌─────────────────────────────────────┐
│   ORCHESTRATOR (Sonnet)             │
│   • Analyzes project                │
│   • Creates work breakdown          │
│   • Classifies tasks                │
│   • Monitors execution              │
└─────────────────┬───────────────────┘
                  │
        ┌─────────┼──────────┐
        ▼         ▼          ▼
    ┌────────┐ ┌──────┐ ┌─────────┐
    │PLANNER │ │ DEV  │ │ DEPLOY  │
    │agents  │ │agents│ │ agents  │
    └────────┘ └──────┘ └─────────┘

Agent Types

| Type | Tools | Use For | Overhead | |------|-------|---------|----------| | PLANNER | Read, Grep, WebSearch | Research, analysis | ~12K tokens | | DEV | Read, Write, Edit, Bash | Implementation, testing | ~10K tokens | | DEPLOY | Bash, Read | Deployment, operations | ~8K tokens |

Execution Example

/c-breakout orchestrate "Implement user authentication"

What happens:

1. Orchestrator Planning (Sonnet agent)

   • Reads your project structure
   • Creates work breakdown: research → implement → test → deploy
   • Classifies 12 tasks across 3 agent types
   • Generates specialized prompts

2. Execution Waves (12 Haiku agents)

   • Wave 1: 3 PLANNER agents research architecture (parallel)
   • Wave 2: 1 DEV agent creates user model
   • Wave 3: 4 DEV agents implement endpoints (parallel)
   • Wave 4: 2 DEV agents write tests (parallel)
   • Wave 5: 2 DEPLOY agents handle migrations & configs (parallel)

3. Results Aggregation

   • Collects all outputs
   • Calculates total cost
   • Generates comprehensive report

Result: Complete auth system implemented in ~51 seconds for $0.48 (vs $1.10 without orchestration)

Commands

In Claude Code

/c-breakout orchestrate <goal>     # Orchestrate project with specialized agents
/c-breakout distribute <task>      # Break down task into sub-tasks
/c-breakout summarize <pattern>    # Summarize files in parallel
/c-breakout tasks                  # Distribute todo list tasks
/c-breakout status [run-id]        # Check execution status
/c-breakout results [run-id]       # Get results from completed run
/c-breakout history                # Show run history
/c-breakout cancel [run-id]        # Cancel running agents

CLI Tool

After installation, you also get a CLI tool for managing runs:

c-breakout status                  # Show latest run status
c-breakout results                 # Show latest results
c-breakout history                 # Show all runs
c-breakout clean                   # Clean up run directories

Cost Comparison

100-Task Project

| Method | Token Overhead | Cost (Haiku) | Savings | |--------|----------------|--------------|---------| | No orchestrator | 1,900,000 | $1.52 | Baseline | | With orchestrator | 1,074,000 | $0.86 | 43.5% |

Why It's Cheaper

• Full context agents: 19K tokens (includes all skills, tools)
• Specialized agents: 8-12K tokens (only tools they need)
• Orchestrator overhead: ~74K tokens (one-time cost)

Example Scenarios

Full Feature Implementation

/c-breakout orchestrate "Add user profile management with CRUD operations"

Plan:

• PLANNER: Research current user model, analyze API patterns
• DEV: Create profile model, implement CRUD endpoints, write tests
• DEPLOY: Run migrations

Cost: ~$0.60 (vs $1.20 without orchestrate)

Refactoring Project

/c-breakout orchestrate "Refactor authentication to use JWT tokens"

Plan:

• PLANNER: Analyze current auth, research JWT best practices
• DEV: Implement JWT service, update endpoints, update tests
• DEPLOY: Update configs, run migrations

Full Stack Feature

/c-breakout orchestrate "Add real-time notifications with WebSocket"

Plan:

• PLANNER: Research WebSocket libraries, analyze architecture
• DEV: Implement WebSocket server, notification service, client handlers, tests
• DEPLOY: Update server configs, deploy with new dependencies

Configuration

Edit ~/.claude/commands/c-breakout.md to customize:

{
  "max_concurrent": 100,
  "orchestrate": {
    "orchestrator_model": "sonnet",
    "agent_model": "haiku",
    "agent_types": {
      "planner": {
        "tools": ["Read", "Grep", "Glob", "WebSearch"]
      },
      "dev": {
        "tools": ["Read", "Write", "Edit", "Bash", "Grep", "Glob"]
      },
      "deploy": {
        "tools": ["Bash", "Read"]
      }
    }
  }
}

When to Use Orchestrate Mode

✅ Use orchestrate for:

• Projects with 10+ tasks
• Full features (planning → development → deployment)
• Multi-file changes
• Projects requiring research + implementation

❌ Don't use orchestrate for:

• Simple single tasks
• Pure calculation tasks
• Quick file summaries

Monitoring Progress

Check status anytime:

/c-breakout status

Output:

=== C-Breakout Status ===
Run ID: 20260131_152030
Mode: orchestrate

Wave 2/3 executing...
Progress: 15/25 agents (60%)

| Agent | Type    | Status      | Task                    |
|-------|---------|-------------|-------------------------|
| P001  | planner | completed ✓ | Research auth patterns  |
| D001  | dev     | running     | Implement JWT service   |
| D002  | dev     | pending     | Write tests             |

Advanced Usage

Custom Agent Classification

Force specific agent types in your goal:

/c-breakout orchestrate "Add caching. Use PLANNER agents to research Redis vs Memcached, then DEV agents to implement."

Adjust Concurrency

For larger projects, edit the config:

{
  "max_concurrent": 200
}

Troubleshooting

"Orchestrator planning failed"

• Make project goal more specific
• Ensure project files are readable
• Fall back to standard distribute mode

"Agent needs blocked tool"

• Agent was misclassified and needs a tool it doesn't have
• Check error message for details
• Manually reclassify or use standard mode

"Dependency deadlock"

• Circular dependencies detected
• Review execution waves
• Manually reorder tasks if needed

Documentation

• Quick Start Guide
• Full Specification
• Examples
• API Reference

Development

# Clone the repository
git clone https://github.com/yourusername/c-breakout
cd c-breakout

# Install dependencies
npm install

# Run tests
npm test

# Publish
npm publish

Contributing

Contributions are welcome! Please read CONTRIBUTING.md for details.

License

MIT © [Your Name]

Support

• 🐛 Report bugs
• 💡 Request features
• 💬 Discussions

Credits

Built for Claude Code by the community.

Ready to orchestrate? Install now and start building with intelligent agent specialization! 🚀