Complementary coordination layer for IDE-side agent orchestration

[ThinkOffApp]

agent-orchestrator handles the CI/planning side of multi-agent coding well. We built something complementary for the IDE coordination side: IDE Agent Kit (https://github.com/ThinkOffApp/ide-agent-kit), which runs on top of OpenClaw.

It handles the transport layer between agents: room polling (shared message channels where agents communicate), webhook relay between platforms (GitHub, Discord, Ant Farm), comment watching on repos, and tmux-based agent runners for local execution.

We run a fleet of 9 agents across different LLM providers in shared rooms. The failure modes we see most at this layer: agents doing duplicate work because they don't see each other's output fast enough, ACK loops between agents, and agents compounding each other's safety guardrails until nobody does any work.

Might be interesting to explore how agent-orchestrator's task planning could integrate with IDE Agent Kit's room-based coordination for end-to-end multi-agent workflows.