How I got Codex and Claude Code to collaborate using their native wake-up mechanisms

Description and use

A lightweight collaboration layer using mailboxes and native wake‑up hooks enables reliable interaction between Claude Code and Codex.

Problem

Integrating two distinct AI coding agents (Claude Code and Codex) so they can collaborate on a single task without interfering with each other’s native workflows.

Outcome

By separating data (mailbox) from control (native wake‑up), the system achieves reliable, asynchronous collaboration between Claude Code and Codex without altering their internal runtimes.

Workflow

• 1. Assign each agent a unique identity in CleanCode.
• 2. Map the identity to the provider’s native runtime (Claude session or Codex thread).
• 3. When Agent A wants to send a message, place it in Agent B’s mailbox instead of injecting it into the prompt.
• 4. Trigger Agent B’s native wake‑up mechanism (e.g., Claude’s FileChanged hook or Codex’s thread notification).
• 5. Agent B wakes, checks its mailbox, processes the message, and replies by placing a response in Agent A’s mailbox.
• 6. Wake Agent A using its native mechanism to deliver the reply.

Evidence

I implemented this as part of CleanCode

Source

The inbox is the data plane. The native wake‑up mechanism is the control plane.