Back to Ecosystem Pulse
ToolExperimentalMCP
cli-agent-orchestrator
by awslabs
Orchestrate coding agents in isolated tmux sessions for replayable multi-agent workflows
Python
Updated Aug 26, 2026
Share:
Summary
Orchestrates multiple AI coding assistants (Claude Code, Kiro, Codex, Copilot CLI, etc.) by running each agent in isolated tmux sessions. Routes prompts and responses between agents to coordinate multi-agent coding workflows and allow side-by-side comparisons. Distinctive for treating each model as an independent CLI actor, preserving session state and terminal logs for reproducible interactions. Tool Use Pattern
Why It Matters
As developers chain specialist coding agents, visibility into how each agent acted and why decisions were made becomes essential for trust. Running agents in isolated tmux sessions produces replayable logs and clear separation of context, enabling inspection of agent behavior and failure modes. That makes it easier to build agent track records, replay interactions for debugging, and integrate lightweight evaluation or reputation layers on top. Agent Registry Pattern
Ideal For
Developers and researchers experimenting with multi-agent coding workflows who need reproducible, inspectable agent interactions and side-by-side model comparisons. Sub-Agent Delegation Pattern Blackboard Pattern
Applications
- Compare output and decision-making between different coding assistants side-by-side
- Replay and inspect agent sessions to diagnose multi-agent system failures
- Stage pre-production tests of agent delegation and interaction patterns before integrating reputation systems
Works With
openaianthropictmux
Topics
agent-orchestrationai-agentsai-coding-assistantantigravityclaude-codecodexcopilot-clicursor-clideveloper-toolshermes+9 more
Similar Tools
autogencrewai
Keywords
multi-agent orchestrationagent delegationagent interaction loggingreproducible agent workflows