This system orchestrates multiple AI agents (Claude Code) to work in parallel, addressing the need for faster, scalable task execution. It targets developers, researchers, and power users who can benefit from an automated, multi-agent workflow, offering parallel processing, non-blocking command input, and persistent memory across sessions.

How It Works

The core architecture employs a hierarchical structure: a "Shogun" agent receives user commands, a "Karo" agent acts as a manager distributing tasks, and up to eight "Ashigaru" agents execute tasks in parallel. Communication is managed via YAML files for structured data exchange and tmux send-keys for event-driven agent wake-ups, eliminating inefficient polling. This design enables rapid delegation, parallel execution, and a non-blocking user experience where new commands can be issued while background tasks complete.

Quick Start & Requirements

• Windows: Download or clone the repository, then run install.bat (as administrator) which handles WSL2 and Ubuntu installation if needed. Follow with first_setup.sh in Ubuntu, and shutsujin_departure.sh for daily operation.
• Linux/Mac: Clone the repository, make scripts executable (chmod +x *.sh), run first_setup.sh, and then shutsujin_departure.sh for daily startup.
• Prerequisites: WSL2 and Ubuntu (required for Windows), tmux, Node.js v20+, and the Claude Code CLI (@anthropic-ai/claude-code). Manual installation guides are provided for dependencies.
• Links: Repository ZIP download: https://github.com/yohey-w/multi-agent-shogun/archive/refs/heads/main.zip. WSL installation: wsl --install (PowerShell).

Highlighted Details

• Parallel Execution: Up to 8 AI workers execute tasks simultaneously, significantly reducing completion times.
• Non-Blocking Workflow: The Shogun agent delegates tasks immediately, returning control to the user without waiting for task completion.
• Memory Across Sessions (Memory MCP): The system maintains persistent memory, allowing AI agents to recall user preferences and past decisions across different sessions.
• Event-Driven Communication: Utilizes YAML files and tmux send-keys for agent communication, avoiding costly API polling.
• Screenshot Support: Integrates with VSCode's Claude Code extension, enabling agents to analyze screenshots provided by the user via a configured path.
• Model Configuration: Differentiates model usage, with Shogun using Opus (disabled thinking) for high-level decisions, and Karo/Ashigaru using Default models for task distribution and execution.

Maintenance & Community

The project is based on "Claude-Code-Communication" by Akira-Papa. No specific community channels (like Discord/Slack) or active contributor details are listed beyond the core dependencies on Anthropic's Claude Code ecosystem.

Licensing & Compatibility

The project is released under the MIT License, which generally permits commercial use and integration into closed-source projects without significant restrictions.

Limitations & Caveats

Windows users must have WSL2 and Ubuntu installed for the system to function. The screenshot analysis feature requires manual configuration of the screenshot save path in config/settings.yaml to match the user's system. Advanced features like "Skills" are not pre-included and must be developed organically based on user-approved candidates. MCP tools require explicit loading before use, and agents must be launched with the --dangerously-skip-permissions flag.