multi-agent-intercom

To send messages between isolated OpenClaw agents across different sandboxes, you can use a cross-boundary peer messaging protocol that calls the native openclaw agent CLI to dispatch work requests to target agents.

The sessions_send tool returns a session not found error because OpenClaw agents run in isolated workspaces, preventing built-in messaging from reaching agents outside the sender's own subagent hierarchy.

To install the agent intercom protocol, ask your agent to run the installation script, which idempotently injects a standard operating procedure into each agent's AGENTS.md file so receivers know how to reply.

Yes, you can wake up a specific OpenClaw agent and force immediate processing by sending instructions that target the main session, prefixing messages with a sender caller ID tag via the openclaw agent CLI.

Multi-agent messaging supports internationalization by detecting the OS language for English or Chinese, and it executes the send operation silently in the background to ensure stealthy asynchronous dispatch.

Available OpenClaw agents are auto-discovered by listing agent IDs from the local openclaw.json configuration, ensuring you target the correct agent when sending cross-boundary messages.