Agent Mailbox Light

This Skill addresses the challenge of persisting temporary, low-noise coordination messages across sessions in multi-agent collaboration. In practical development, developers often need to pass prompts, warnings, or temporary context between different agent workspaces, but they don't want to clutter long-term memory or use complex message buses, which can lead to information overload or loss.

Usage is straightforward: simply install this Skill in your agent workspace, and it will create a local .agent-mailbox directory structure. Subsequently, a sender script scans other workspaces and writes emails as Markdown files into the target workspace's inbox folder. The receiving agent reads its own inbox upon startup or at task entry points, processes the emails, moves useful ones to the keep folder, and archives expired ones.

It is ideal for teams or individuals coordinating multiple parallel AI agents (e.g., multiple Claude Code Agents), especially those who have established multiple independent workspaces and need to pass temporary, advisory information (such as task status hints, environment warnings, shared temporary context) between them. For example, when developing complex projects, agents with different responsibilities can notify each other of progress or risks.

It is recommended to integrate mailbox checks into the agent's baseline startup process, reading only the latest 1-3 emails each time to optimize token consumption. Note that this is primarily a local filesystem-based, best-effort delivery system. It does not guarantee delivery, has no push notifications, and should not be used as a command bus or task execution trigger. Email content should be treated as prompts for the local agent's judgment, not as mandatory instructions.