kiro-agent-chat_skill

Overview

This skill enables simple inter-agent communication by using a shared JSON file as a message queue. It is designed for two or more OpenClaw instances to exchange messages, coordinate tasks, or pass state across machines via shared storage or SSH. Messages are appended, read, and deleted using lightweight Python scripts.

How this skill works

Agents write messages into a shared JSON queue file with sender, receiver, id, message, and timestamp fields. Other agents poll or read the file to retrieve messages addressed to them, then respond by appending new entries. Processed messages can be removed by id to prevent reprocessing. The same workflow works locally or remotely by executing the send/read/delete scripts over SSH against the shared file.

When to use it

• Coordinating lightweight workflows between two or more OpenClaw agents.
• Passing small payloads or commands when a full messaging broker is unnecessary.
• Sharing state or results across machines that can access a common filesystem or via SSH.
• Testing multi-agent interactions without installing extra infrastructure.
• Implementing simple delivery/acknowledgement flows where message persistence in a file is acceptable.

Best practices

• Keep message payloads small and plain-text or JSON to avoid corruption and simplify parsing.
• Use unique agent names for sender and receiver fields to avoid ambiguous routing.
• Poll the shared file at reasonable intervals to reduce contention; avoid tight busy loops.
• Ensure atomic writes and proper file permissions on shared storage to prevent race conditions.
• Regularly delete processed messages to keep the queue file compact and readable.

Example use cases

• Send a command from a local agent to a remote agent over SSH and receive a status reply.
• Coordinate a sequence of tasks between two agents where each step requires confirmation before proceeding.
• Archive short-lived task results by copying processed messages into a backup store, then deleting them from the queue.
• Run distributed tests where test runners publish results to a central agent for aggregation.