[GPT-5.6 SOL MAX–generated diagnostics] Codex Micro intermittently freezes/disconnects while wired over USB

Document type: AI-generated diagnostics one-pager Generated by: OpenAI Codex Model: GPT-5.6 Sol (gpt-5.6-sol) Reasoning effort: Max Generated: July 18, 2026, America/Los_Angeles Evidence: User observations, Codex device logs, and macOS Bluetooth/USB/HID events inspected during live troubleshooting Privacy: Local filesystem paths are intentionally omitted. Full diagnostic files were shared through Codex feedback. Feedback ID: 019f7796-8019-7ce0-bd3a-1b373949321b

TL;DR

Codex Micro experiences unpredictable, intermittent freezes or disconnects while physically connected over USB.

It may work for approximately a second and then stop responding for several seconds. Switching between Codex tabs or tasks sometimes appears to make the problem worse, but freezes also happen without any user interaction. Sometimes the keyboard recovers automatically; sometimes it never recovers and requires reconnection or another recovery action.

The device is physically connected over USB, but Bluetooth remains paired. Logs show real HID read failures and Bluetooth disconnect/reconnect activity. The sequence is consistent with a possible Bluetooth-to-USB fallback, transport handoff, or stale transport state, but the current logs do not explicitly identify which transport Codex selected for each operation.

A separate captured freeze involved an RGB configuration RPC timing out after exactly ten seconds, even though macOS recorded no corresponding USB detach, reset, or Bluetooth transport switch. This suggests that the problem may also involve the Codex device communication layer rather than only physical connectivity.

Environment

• Codex desktop version: 26.715.31925
• Codex build: 5551
• macOS version: 26.5.2
• macOS build: 25F84
• Codex Micro firmware: v0.4.1
• USB vendor/product ID: 0x303A:0x8360
• Bluetooth pairing name: Codex Micro #1
• Device profile: 0
• Active layer: 1
• Battery: approximately 80–82%
• Charging status: charging
• Physical connection during investigation: USB
• Bluetooth state: paired and available

User-visible symptoms

• Input intermittently freezes for several seconds.
• The device may work for approximately one second and then freeze again.
• Freezing sometimes becomes more frequent when switching between Codex tabs or tasks.
• Freezing also occurs without tab switching or other deliberate user interaction.
• Recovery is inconsistent:
• Sometimes the device resumes after a few seconds.
• Sometimes the interruption lasts tens of seconds.
• Sometimes it never recovers without intervention.
• The behavior is random and difficult to reproduce reliably.
• The device can remain visibly connected and charging while communication is stalled.
• There is not always an obvious macOS USB-disconnection event when a freeze occurs.

Captured failures

All times below are in PDT.

HID read failure associated with Bluetooth disconnect

At approximately 17:49:57.189, macOS recorded a Bluetooth Low Energy disconnect for Codex Micro.

At 17:49:57.237, Codex reported that it could not read from the HID device.

The device reconnected approximately 11 seconds later.

This ordering could indicate a Bluetooth failure followed by USB fallback or transport reinitialization. However, the application logs do not expose the selected transport clearly enough to confirm that interpretation.

Extended HID interruption

At 18:01:14.830, Codex recorded another HID read failure.

Communication resumed at approximately 18:02:25.885, producing an interruption of about 71 seconds.

Additional HID interruption

At 18:02:38.317, another HID read failure occurred.

Communication resumed at approximately 18:02:59.341, approximately 21 seconds later.

Ten-second RGB RPC timeout

At 18:10:24.877, Codex sent an RGB configuration request:

v.oai.rgbcfg
request id: 90

Exactly ten seconds later, the request failed with:

WLDeviceError
TIMEOUT

Other device requests resumed immediately afterward.

No matching USB detach, USB reset, or Bluetooth transport switch was found in the corresponding macOS event window. This incident looks more like a stalled RPC or device-communication operation than a physical cable disconnect.

Focus-related warnings

The logs also contained repeated warnings similar to:

No resolver found for host.focused_app

These were observed around:

• 18:07:03
• 18:09:46
• 18:12:03

These warnings may be unrelated. However, because switching Codex tabs or tasks sometimes appears to worsen the freezing, they may be useful when investigating focus-driven device updates or application-state changes.

Transport observations

• The device was physically connected and detected through USB during inspection.
• Bluetooth remained paired as Codex Micro #1.
• At the latest check, the active HID device appeared to be USB.
• No active Codex Micro Bluetooth HID connection was shown at that time.
• Earlier logs nevertheless contained Bluetooth disconnect/reconnect activity close to HID read failures.
• The available logs do not explicitly state:
• which transport handled each individual request;
• whether Codex automatically switched between Bluetooth and USB;
• whether both transports were briefly active;
• whether a stale Bluetooth connection blocked USB communication;
• whether USB was explicitly preferred because the device was wired.

Automatic Bluetooth switching or USB/Bluetooth contention is therefore a plausible hypothesis, not a confirmed conclusion.

Sampled reliability

In one inspected ten-minute window:

• 133 successful device responses
• 0 HID write errors
• 2 HID read drops
• 0 device-open errors
• 2 transport-related warnings

This matches the user experience: most operations succeed, but sporadic failures produce noticeable and sometimes prolonged freezes.

Other software and prior troubleshooting

Karabiner-Elements previously attempted to grab the Codex Micro keyboard. An explicit ignore rule is now present and remained active during the latest failures:

{
  "identifiers": {
    "is_keyboard": true,
    "product_id": 33632,
    "vendor_id": 12346
  },
  "ignore": true
}

The current recurrence therefore appears separate from the earlier Karabiner conflict.

A previous nonrecovering incident was cleared only after logging out of and back into macOS. This suggests that stale HID or transport state may sometimes persist outside the Codex process.

Expected behavior

When Codex Micro is connected over USB:

• USB should remain the stable and preferred transport.
• A paired Bluetooth connection should not interrupt USB communication.
• Switching Codex tabs or tasks should not freeze device input.
• Device configuration RPCs should complete promptly or fail without blocking input.
• A failed request or transport should recover automatically and predictably.
• Recovery should not require reconnecting the device or logging out of macOS.

Actual behavior

• Input and device communication freeze intermittently.
• HID reads sometimes fail.
• Interruptions range from a few seconds to more than one minute.
• Some freezes do not recover automatically.
• Bluetooth activity occasionally occurs close to HID failures despite the physical USB connection.
• At least one ten-second RPC timeout occurred without a corresponding physical transport event.
• Switching tabs or tasks sometimes correlates with increased freezing.
• The same problem also occurs spontaneously without user interaction.

Requested investigation

Please investigate:

1. How Codex chooses between USB and Bluetooth when both are available.
2. Whether USB is explicitly preferred while Codex Micro is wired.
3. Whether a paired Bluetooth transport can cause USB transport reinitialization or contention.
4. Whether both transports can become active or partially active simultaneously.
5. Whether failed HID reads leave stale transport state behind.
6. Whether v.oai.rgbcfg or other configuration RPCs can block the device communication queue.
7. Whether tab or task switching triggers focus, profile, layer, or RGB requests that can stall input.
8. Whether the host.focused_app resolver warnings are connected to device updates triggered by focus changes.
9. Whether transport selection and handoff decisions can be added to diagnostic logs.
10. Whether RPC timeout logs can include the active transport, endpoint, queue state, and attempted recovery action.
11. Whether Codex can automatically reset only the affected device session instead of requiring reconnection or a macOS logout.

Suggested reproduction procedure

Because the failure is intermittent:

1. Pair Codex Micro over Bluetooth.
2. Connect it physically through USB while leaving Bluetooth enabled.
3. Use the device normally in Codex.
4. Switch repeatedly between Codex tabs or tasks.
5. Leave the device idle periodically to capture spontaneous freezes.
6. Monitor HID reads, active transport selection, transport handoffs, focus-related updates, and configuration RPCs.
7. Continue testing for at least 10–20 minutes.
8. Test the same workflow again after explicitly disabling Bluetooth to compare failure rates.