This email is a backup record of a critical Codex Desktop crash diagnosis (2026-08-11, Asia/Shanghai). The same content is being submitted as a GitHub issue to GitHub - openai/codex: Lightweight coding agent that runs in your terminal · GitHub.

CRITICAL: Codex Desktop App — Recurring Browser-Main-Process Crash (30+ times/day, blocks production work)

Severity: CRITICAL (P0) — production-blocking, repeated data/task loss, no user-side workaround
Date: 2026-08-11 (Asia/Shanghai)
Affected build: OpenAI.Codex_26.803.10989.0_x64__2p2nqsd0c76g0 (Chromium 151.0.7922.76)

1. Summary

The Codex desktop app crashes deterministically and repeatedly. On 2026-08-11 the app crashed 30+ times between 13:15 and 18:34 (about 5.5 hours). Every crash is the SAME browser main process fatal breakpoint:

• Exception: 0x80000003 (STATUS_BREAKPOINT / break instruction)
• Faulting module: chrome.dll at offset +0x9362037
• Faulting thread: CrBrowserMain (browser main thread), dispatching a window message (USER32!DispatchMessageWorker → USER32!UserCallWinProcCheckWow → V8/Blink code path)

This is an in-app deterministic crash caused by the app’s own code — NOT an external kill, NOT WER, NOT a GPU process crash, NOT hardware acceleration, and NOT a system/OS fault.

2. Evidence — three independent captures, identical crash site

All three land on the exact same instruction. The two live-debugger call stacks are frame-for-frame identical.

Module identity (for build lookup)

• chrome.dll: File version 151.0.7922.76, ProductName “Codex” (custom OpenAI build)
• Base 0x00007ffc20840000, ImageSize 0x137F3000, Timestamp 0x6A711C95 (2026-08-04)
• PDB: GUID {36A5B6A2-61F6-D72C-4C4C-44205044422E}, Age 1, chrome.dll.pdb
• The PDB is NOT on the public Microsoft symbol server (HTTP 404) — OpenAI must resolve with internal symbols.

Crash-site disassembly (captured live at 18:33:20)

00007ffc29ba2017  jne   ...          ; conditional branch
00007ffc29ba202a  mov   rcx,[rsp+38h]  ; stack-cookie load
00007ffc29ba202f  xor   rcx,rsp        ; stack-cookie check
00007ffc29ba2032  call  chrome+0x5F28780  ; noreturn fail-fast / CHECK path
00007ffc29ba2037  int   3             ; <-- CRASH (0x80000003)
00007ffc29ba2038  ud2                 ; unreachable marker

3. Reproduction & blast radius

• Crashes occur on instances launched with the default user-data-dir (no --user-data-dir flag).
• A differently-profiled instance running the same tasks all day with --user-data-dir=Codex-DeepSeek had ZERO crashes (its Crashpad DB contains only a 2026-08-09 record). This isolates the bug to profile/session-state related data or code paths, not the GPU/system.
• Time-to-crash is random: observed 50 seconds to 69 minutes after instance start.
• Crashes happen while tasks are actively running (thread polling, turn-state updates). Immediately before crashes, the overlay renderer was observed flooding errors: Conversation state not found, Received item/* for unknown conversation, plus recurring ResizeObserver loop completed with undelivered notifications and Git snapshot became stale.
• After a crash the app auto-restarts, then crashes again within minutes → denial-of-service loop.

4. Mitigations attempted (all failed except profile switch)

• Disabled hardware acceleration (hardware_acceleration_mode: 2) → still crashed 4 minutes later (17:08:54), same signature.
• App updated today (09:46 store auto-update + 14:30 rebuild) → crashes on both old and new builds.
• No WER / Application Error / AppX crash events are produced → silent in-app self-termination.

5. Related reports (same family: overlay renderer + ResizeObserver + controlled process-tree exit)

• GitHub #33996 — concurrent reasoning summaries correlate with ResizeObserver loops and UI freezes (hidden avatarOverlay receives duplicated events).
• GitHub #33561 — severe lag + repeated 0xc06d007f; hidden avatarOverlay in 85 log entries, repeated ResizeObserver errors.
• community.openai.com — “Codex Desktop for Windows Repeatedly Exits and Relaunches During Reasoning Summaries” — whole process tree exits with code 0, relaunches 5-8s later, no WER/Crashpad, ResizeObserver flood before exit.

6. What we need from OpenAI

1. Resolve chrome.dll+0x9362037 against build PDB GUID {36A5B6A2-61F6-D72C-4C4C-44205044422E} and identify the failing CHECK and its trigger condition.
2. Prioritize a hotfix/patch release; this is a P0 production blocker.
3. Investigate the overlay / turn-state path as a suspected trigger (it floods Conversation state not found errors immediately before crashes).
4. Fix the auto-restart loop so a crashing profile does not immediately relaunch into the same crash.

7. Contact expectation

Please treat this as P0. We expect: (a) root-cause confirmation within 48h using the PDB GUID above, (b) a patched build, and (c) an escalation path for crash evidence upload.

— Affected user, 2026-08-11