o oopunsoosu
v1.0.0 · MIT License

Multi-Agent Parallel
Development Framework

Coordinate multiple Claude Code agents through tmux sessions and YAML-driven task queues. A hierarchical command structure — Shogun, Karo, Ashigaru — keeps agents focused, parallel, and conflict-free.

# Command hierarchy

         ┌──────────┐
         │  SHOGUN  │  ← Project orchestrator
         └────┬─────┘
       YAML task queues
         ┌────▼─────┐
         │   KARO   │  ← Task manager & distributor
         └────┬─────┘
       YAML task assignments
  ┌───┬───┬──▼┬───┬───┬───┬───┐
  │A1 │A2 │A3 │A4 │A5 │A6 │A7 │A8 │  ← Ashigaru workers
  └───┴───┴───┴───┴───┴───┴───┴───┘

Features

What makes oopunsoosu different from running agents in separate terminals.

Parallel Agent Execution

Run up to 8 agents simultaneously in tmux panes. Each agent operates in its own pane with dedicated task files, eliminating write conflicts between workers.

YAML-Driven Task Queues

Tasks are defined and tracked as YAML files. Shogun writes to a queue, Karo distributes to per-worker files, Ashigaru reads and reports back. No polling — event-driven via tmux send-keys.

Native tmux Integration

Built on tmux sessions with named windows and indexed panes. Monitor all agents at a glance, inspect output in real time, and intervene when needed — all from one terminal.

Hierarchical Command Structure

Inspired by feudal military organization. Shogun sets strategy, Karo manages task distribution and quality, Ashigaru execute. Clear chain of command prevents agents from stepping on each other.

Getting Started

Three steps to your first multi-agent session.

1

Clone the repository

$ git clone https://github.com/kawagutt/multi-agent-shogun.git
$ cd multi-agent-shogun
2

Configure your settings

$ cp config/settings.yaml.example config/settings.yaml
$ vim config/settings.yaml  # Set language, agent count, etc.
3

Launch the agents

$ bash shutsujin_departure.sh
# Spawns tmux sessions: shogun (1 pane) + multiagent (up to 9 panes)
# Shogun, Karo, and Ashigaru agents start automatically
Prerequisites: macOS or Linux, tmux, Claude Code CLI, and a Claude API key.

How It Works

The communication flow between agents.

queue/shogun_to_karo.yaml

Shogun writes high-level objectives. Karo picks them up and breaks them into subtasks for individual Ashigaru.

queue/tasks/ashigaru{N}.yaml

Each Ashigaru has a dedicated task file. No shared writes, no race conditions. Karo notifies workers via tmux send-keys.

queue/reports/ashigaru{N}_report.yaml

Workers report back through dedicated report files. Karo aggregates results and updates the dashboard for human review.