Agent Mailbox Light

This Skill addresses the need for temporary, low-noise coordination between multiple AI agent workspaces without writing information to long-term memory. In practical development, developers often need to pass prompts, warnings, or context between multiple parallel agent projects but do not want these temporary messages to pollute core memory or trigger complex task scheduling. Agent Mailbox Light offers a lightweight, filesystem-based mailbox, allowing agents to view notifications from other workspaces—much like checking email—upon startup or during task intervals.

Usage is straightforward: simply install this Skill in your agent workspace and configure the path pattern to scan other workspaces (e.g., $HOME/.openclaw/workspace*). Subsequently, sender scripts will automatically write mail files to the .agent-mailbox/inbox/ directory under matching paths, while recipient agents can read the latest mail in their own inbox as part of their baseline workflow. After summarizing useful information, they can move it to the keep folder and archive processed or expired mail.

It is ideal for teams or individuals coordinating multi-agent tasks, especially developers already using multiple independent workspaces (e.g., one for front‑end development, one for data analysis, one for operations monitoring) who need a simple, dependency‑free mechanism to share temporary context, warnings, or task‑status updates without introducing complex message queues or centralized services.

We recommend integrating mailbox‑checking logic uniformly into the agent’s startup or task‑switching baseline—for instance, reading only the latest 1–3 messages per startup to control token consumption. Note that it is primarily a “best‑effort” broadcast system: it does not guarantee delivery, provides no push notifications, and mail content should be treated as “suggestions” for the local agent to evaluate, not as commands that must be executed.