Claude Swarm

Claude Swarm

Multi-agent orchestration for Claude Code — decompose complex tasks into parallel subtasks, coordinate agents in real-time, and visualize everything in a rich terminal UI.

Built with the Claude Agent SDK for the Claude Code Hackathon (Feb 10-16, 2026).

How It Works

1. Task Decomposition — Describe a complex task. Opus 4.6 analyzes your codebase and breaks it into a dependency graph of subtasks
2. Parallel Agent Spawning — Independent subtasks run simultaneously via Claude Agent SDK. Dependent tasks wait.
3. Real-time Coordination — File conflict detection prevents agents from stepping on each other. Budget enforcement stops runaway costs.
4. Opus Quality Gate — After agents complete, Opus 4.6 reviews the combined output for correctness, consistency, and completeness
5. Rich Terminal UI — htop-style dashboard showing agent progress, tool usage, costs, and file conflicts in real-time
6. Session Replay — Every swarm execution is recorded. Replay any session to review what each agent did.

Quick Start

Architecture

Features

CLI Reference

YAML Configuration

Create swarm.yaml in your project root to define custom agent types:

Claude Swarm auto-detects swarm.yaml or .claude/swarm.yaml in your project.

How Opus 4.6 Is Used

Claude Swarm demonstrates strategic model selection with two critical Opus 4.6 touchpoints:

Phase 1: Task Decomposition (Planning)
Opus 4.6 handles the hardest reasoning task — analyzing your codebase, understanding the architecture, identifying dependencies between subtasks, and producing a parallelizable execution plan. This requires deep understanding of code relationships and optimal task splitting.

Phase 2: Worker Execution
Haiku handles the parallelizable work — each agent follows focused instructions from the plan. Using…