agent-mailbox

Coordinate multiple AI agents through append-only Markdown messages in a shared folder.

14|1|Updated Jun 14, 2026
One-click install
npx skills add https://github.com/allemaar/open-skills --skill agent-mailbox-allemaar
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: agent-mailbox
Source: https://github.com/allemaar/open-skills/tree/main/skills/agent-mailbox
Command: npx skills add https://github.com/allemaar/open-skills --skill agent-mailbox-allemaar

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve? Getting two or more AI agents—across different runtimes and vendors—to collaborate reliably is hard: chat transcripts are hidden in one runtime, messages get missed, and there is no auditable record of who did what. This Skill turns any shared folder into a structured mailbox where agents handshake, exchange causal messages, deliver artifacts, and recover from missed messages. ## Core Features & Use Cases - Folder-based agent messaging: Agents exchange append-only Markdown envelopes with canonical metadata (UUIDs, causal parents, recipients) over local folders, Git, OneDrive, or Lyt vaults—no server or daemon required. - Durable disposition tracking: Every addressed message receives an append-only, UUID-keyed handling record, so restarts and missed messages are reconciled instead of lost. - Bounded listening and recovery: Optional participant-local listener and scheduled-check packages with deadlines, failure budgets, and honest PARKED/DEGRADED states, plus a missed-message recovery protocol. - Use Case: Have a Claude Code agent and a Codex agent collaborate on a code review: one delivers an artifact with its hash into the shared inbox, the other reconciles its inbox, reviews the artifact, and replies with blocking defects—all auditable in the folder. ## Quick Start Share a folder between two agents, load this skill in each, and tell your agent to handshake with the other agent through that mailbox folder.

Frequently Asked Questions about agent-mailbox

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I make two AI agents collaborate through a shared folder?

Point both agents at the same mailbox folder and load this skill in each. They perform a two-message handshake, then exchange append-only Markdown envelopes with causal UUIDs for delivery, review, and status messages.

Can agents on different platforms like Claude Code and Codex communicate?

Yes, the protocol is runtime- and vendor-agnostic because the transport is just a shared folder. Claude Code can use a native Monitor-based listener while Codex uses bounded scheduled checks, and both reconcile the same inbox.

What transports does the agent mailbox support?

It works over a local folder, a Git repository, a registered Lyt vault, OneDrive, or other sync-share folders. Each route separates publication, materialization, detection, and task wake so failures can be diagnosed per layer.

What happens when an agent misses a message?

The participant marks itself DEGRADED, stops the faulty listener, and runs the missed-message-recovery protocol: an age-independent reconciliation of the whole addressed inbox against durable UUID dispositions. Readiness is restored only after a fresh real message proves the full path.

Does the mailbox require a server, daemon, or database?

No, it is an operating protocol over plain Markdown files in a folder the user controls. Dispositions and cursors live in participant-local state outside the sync folder, and no broker or queue is installed.

When should I use a one-shot handoff instead of an agent mailbox?

Use a one-shot transfer when there is no ongoing collaboration—just a single artifact or context pass between agents. The mailbox is designed for sustained multi-message exchanges with reconciliation, claims, and recovery.