Agent Mailbox Light

This Skill addresses the need for low-noise, temporary coordination and information sharing among multiple AI agents across different workspaces. In practical development, agents often need to pass prompts, warnings, or context between different tasks or sessions, but they may not want to write these temporary messages into long-term memory or rely on complex message buses, which can lead to high coordination overhead and information clutter.

Usage is straightforward: simply enable this Skill in the agent's workspace, and it will create a .agent-mailbox directory under each workspace, containing three folders: inbox, keep, and archive. The sender script scans other workspaces based on configured path patterns and writes emails as Markdown files into the target workspace's inbox folder. The receiving agent can read the latest emails from its own inbox upon startup or before executing a task, quickly obtaining relevant context, and move emails to the keep or archive folders as needed.

It is ideal for developers or teams who need to coordinate multiple AI agents working in parallel (e.g., multiple Claude Code agents), especially those already using frameworks like OpenClaw and needing to pass temporary prompts, status updates, or warnings between agents across different projects or sessions, while keeping the system lightweight, local, and free of complex dependencies.

It is recommended to integrate mailbox checks into the agent's baseline startup routine, such as automatically reading the latest 1-3 emails each time it starts. Note that this is primarily a "best-effort" broadcast system based on the local file system; it does not guarantee delivery, has no push notifications, and is not suitable as a task execution bus. Email content should be treated as advisory prompts for the local agent to consider, not as mandatory commands.