How jedward23/tmux-orchestrator works
Project type and purpose
• This is a CLI-based orchestration tool that runs multiple Claude AI agents inside tmux sessions.
• It provides autonomous agent scheduling, coordination, persistence, and Git backup for multi-agent workflows.
File structure
• Top-level scripts:
• send-claude-message.sh – sends messages to any Claude agent window
• schedule_with_note.sh – schedules future check-ins with notes
• tmux_utils.py – Python utilities for tmux session/window discovery
• tmux orchestration helper files (.claude/commands/*.md, CLAUDE.md)
• README.md, LEARNINGS.md – documentation and knowledge base
• Examples/ – sample screenshots of orchestrator, PMs, and engineers in action
• .pycache/ – compiled tmux_utils cache
Extract architecture insights from README
• Three-tier agent hierarchy:
1. Orchestrator agent (root coordinator)
2. Multiple Project Manager agents
3. Multiple Engineer agents under each PM
• Core technologies:
• tmux for session/window management and persistence
• Bash scripts for scheduling and messaging
• Python helper library for tmux introspection
• Git for automatic backups and versioning
• Autonomous features: self-triggered check-ins, cross-project coordination, parallelism
Detail capture
• That .claude/commands defines custom agent behaviors
• That CLAUDE.md and LEARNINGS.md are static knowledge sources consulted by agents
• Examples/ as a demonstration artifact, not part of runtime