Desktop remote control intermittently unavailable after network changes or sleep

Description

Codex Desktop remote access is intermittently unavailable when I am away from home. The local app appears to start the remote-control WebSocket, but the connection is later closed and does not reliably reconnect.

Environment

• macOS desktop app
• Codex Desktop version: 26.707.51957
• Installation has a stable remote-control installation ID
• Observed in logs through 2026-07-14

Evidence

The local logs repeatedly show remote control starting with:

desired_state=Unknown

and then:

app-server remote control websocket loop started

The latest observed disconnect is:

forwarding remote control connection closed transport event connection_id=ConnectionId(69)

There is no corresponding actionable reconnect, backoff, or network error logged afterward.

Normal API connectivity appears healthy: HTTPS requests to chatgpt.com succeed, and Codex response WebSockets successfully upgrade to wss://chatgpt.com/backend-api/codex/responses. This suggests the issue is specific to the persistent remote-control channel rather than general outbound connectivity.

Expected behavior

After temporary network loss, Wi-Fi changes, sleep/wake, or app restart, the remote-control WebSocket should reconnect automatically and publish a clear connected/available state.

Actual behavior

Remote access becomes unavailable while the desktop app is still running. The logs show an unknown desired state and a closed remote-control connection without a useful reconnect/error explanation.

Suggested investigation

• Why does remote control initialize with desired_state=Unknown?
• Does a closed remote-control WebSocket restart with exponential backoff?
• Is the desired state persisted and restored after sleep/network changes?
• Add explicit close reasons and reconnect attempts to the logs.
• Verify remote-control state transitions are emitted after reconnect.

I can provide sanitized local log excerpts if needed.