Summary

Add a first-class handoff flow so a user can continue an existing local Codex session inside OpenClaw, then let OpenClaw keep driving it (including async follow-ups/pings) after the user disconnects.

Problem

Today we can spawn ACP sessions from OpenClaw, but we cannot cleanly "adopt" an already-running local Codex session and continue it through OpenClaw channels/threads.

This breaks the handoff loop:

• user starts work locally in Codex
• user needs to leave
• user wants OpenClaw to take over that same session, continue work, and notify/ping if needed

Desired UX

• "Continue my recent Codex session about X here"
• OpenClaw finds/selects the correct resumable Codex session
• OpenClaw binds that session to the current conversation/thread
• Future messages route to that same underlying session
• Agent can continue autonomously and post updates back in-channel

Scope

• ACP/acpx-backed session import/adoption for Codex (and extensible to other ACP harnesses)
• Session identity mapping so OpenClaw can reference the adopted session safely
• Clear lifecycle controls (adopt, focus/unfocus, detach, close)
• Safe behavior across restarts/crashes

Acceptance Criteria

• User can adopt an existing Codex session into OpenClaw without losing context
• Adopted session can be bound to a thread and used multi-turn
• Session survives OpenClaw restart and remains routable
• User can detach or close without corrupting source session state
• Failure modes are explicit (session not found, auth mismatch, busy session, etc.)

Notes

This would complete the local-to-agent handoff loop and unlock a practical "I have to leave, keep going and ping me" workflow.