Codex Desktop renderer crashes after subagent wait with React #185 and unknown conversation state

[malikdraz]

What version of the Codex App are you using (From "About Codex" dialog)?

Codex App 26.623.101652, build 4674

Local evidence:

• /Applications/Codex.app/Contents/Info.plist
• CFBundleShortVersionString = 26.623.101652
• CFBundleVersion = 4674
• Sentry release: codex@26.623.101652

What subscription do you have?

ChatGPT Pro

What platform is your computer?

Darwin 25.5.0 arm64 arm

Additional local context:

• macOS 26.5.1, build 25F80
• app arch: arm64

What issue are you seeing?

Codex Desktop shows the full-screen error boundary:

Oops, an error has occurred

with the buttons:

Update Codex and Try again

The app is already up to date, so the Update Codex action is misleading here. The failure appears to be renderer/session-state related, not an update problem.

This happened while reopening/inspecting a local Codex Desktop thread that had spawned multiple subagents. The affected parent thread id was:

019f2bac-7b89-7002-b960-fc3639bd5a3c

The relevant subagent ids were:

• 019f2bd9-8906-7122-ac08-79cc2a0912a2
• 019f2bd9-a7f7-7890-b852-68e142cf8a2a
• 019f2bd9-c835-7893-b66f-0f0e6d9fabbf

Two subagents returned completion notifications. The third subagent stayed unresponsive through repeated wait_agent calls and was later closed. Around the same window, the renderer logged repeated unknown-conversation/state errors and then React error #185.

Local Sentry breadcrumb counts from ~/Library/Application Support/Codex/sentry/scope_v3.json:

• 10 occurrences of Minified React error #185
• 69 breadcrumbs matching unknown conversation or missing conversation state

Sample React error:

2026-07-04T06:40:47Z
Error: Minified React error #185; visit https://react.dev/errors/185 for the full message or use the non-minified dev environment for full errors and additional helpful warnings.
    at _e (app://-/assets/local-conversation-thread-CRryh-25.js:8:72757)
    at a (app://-/assets/app-initial~app-main~worktree-init-v2-page~remote-conversation-page~onboarding-page~hotkey-~ke3yc5wu-BLQiF1Gs.js:173:21245)

Sample unknown-conversation/state breadcrumbs:

2026-07-04T07:07:46Z [electron-message-handler] Received item/completed for unknown conversation conversationId=019f2bf2-f17b-7081-bb03-7a3f68e6b27b
2026-07-04T07:07:46Z [electron-message-handler] Conversation state not found conversationId=019f2bf2-f17b-7081-bb03-7a3f68e6b27b
2026-07-04T07:07:48Z [electron-message-handler] Received item/started for unknown conversation conversationId=019f2bf2-f17b-7081-bb03-7a3f68e6b27b

What steps can reproduce the bug?

I do not have a minimal deterministic repro yet, but this is the observed shape:

1. Open Codex Desktop 26.623.101652 on macOS.
2. Use a local thread that dispatches multiple subagents.
3. In this case, the parent thread spawned three multi_agent_v1 explorer agents.
4. Wait for subagent results with wait_agent.
5. Two subagents complete and send <subagent_notification> messages.
6. One subagent remains unresponsive through repeated waits.
7. Reopen or navigate around the affected local conversation while this state exists.
8. The renderer eventually shows the full-screen Oops, an error has occurred fallback with Update Codex and Try again.
9. Local breadcrumbs show repeated unknown-conversation / missing conversation state errors and React error feat: experimental side-by-side diff mode in single-pass (WIP) #185 in local-conversation-thread.

Relevant parent-thread lifecycle:

2026-07-04T06:38:24Z spawn_agent -> 019f2bd9-8906-7122-ac08-79cc2a0912a2
2026-07-04T06:38:32Z spawn_agent -> 019f2bd9-a7f7-7890-b852-68e142cf8a2a
2026-07-04T06:38:40Z spawn_agent -> 019f2bd9-c835-7893-b66f-0f0e6d9fabbf
2026-07-04T06:39:16Z wait_agent on all three, timeout 240000ms
2026-07-04T06:40:51Z wait_agent on remaining two, timeout 240000ms
2026-07-04T06:42:08Z wait_agent on remaining one, timeout 240000ms
2026-07-04T06:49:17Z wait_agent on remaining one, timeout 60000ms
2026-07-04T06:51:12Z close_agent on all three

The third sidecar session file existed but only had four JSONL lines and did not produce a normal assistant/tool response before shutdown.

What is the expected behavior?

Codex Desktop should keep the parent thread renderable even if a subagent stalls, emits late notifications, or has an incomplete sidecar session.

More specifically:

• A stalled or closed subagent should not poison the parent LocalConversationThread render path.
• Events for unknown conversation ids should be ignored, quarantined, or attached only after the conversation is registered.
• React feat: experimental side-by-side diff mode in single-pass (WIP) #185 should not reach the app-wide error boundary.
• The full-screen fallback should not suggest Update Codex when the installed app is already current.
• Try again should recover the view or provide a useful diagnostic instead of looping back to the same crash.

Additional information

This looks related to, but not exactly the same as, several existing reports:

• Oops, an error has occurred #23174
• Codex Desktop reconnect loop on new/restarted conversations due to turn/started for unknown conversation #21167
• Codex App/CLI: subagent tasks trigger reconnect loop and stream disconnect #24475
• Renderer repeatedly crashes to "Oops, an error has occurred" — unhandled CancelledError from app-server-manager-signals query invalidation #29061
• Windows Desktop: switching sandbox from read-only to workspace-write crashes UI after world-writable ACL scan #19238
• Codex Desktop subagent completions render as raw notifications and create unknown-conversation errors #23821

The closest overlap appears to be the combination of:

• subagent/session state corruption
• unknown conversation routing
• local conversation renderer crash
• misleading full-screen Update Codex / Try again fallback

I can provide a redacted screenshot of the fallback screen and sanitized local breadcrumb snippets if useful.

[github-actions]

Potential duplicates detected. Please review them and close your issue if it is a duplicate.

• close_agent can hang indefinitely on previously interrupted subagent, blocking parent thread #31036
• Bound v1 close_agent shutdown waits for unresponsive child agents #29937

Powered by Codex Action

[AliHakanVesca]

Adding another reproduction on a newer Codex Desktop build.

Environment:

• Codex app: 26.623.141536, build 4753
• Bundled CLI: codex-cli 0.142.5
• Release: codex@26.623.141536
• OS: macOS 26.3 build 25D125
• Architecture: arm64
• Timezone: Europe/Istanbul

Observed issue:

When opening an existing local conversation/thread in Codex Desktop, the UI switches to the full-screen error boundary:

Oops, an error has occurred
[Update Codex] [Try again]

Try again does not reliably recover the affected conversation. The app already appears up to date, so Update Codex is not actionable in this case.

Observed timing:

• User-visible failure occurred around 2026-07-07 14:25-14:28 +03.
• App logs for the same window are around 2026-07-07T11:25-11:28Z.

Redacted local evidence:

• Sentry scope currently includes 3 breadcrumbs with Minified React error #185.
• Sentry scope currently includes 5 breadcrumbs matching unknown conversation.
• App logs show AppRoutes hitting an error boundary with stack frames including:
  • local-conversation-page-CfMUdpO1.js
  • local-conversation-thread-DhV7q0nP.js
• App logs near the visible failure show repeated unknown conversation events:

2026-07-07T11:25:25Z error [electron-message-handler] Received turn/started for unknown conversation conversationId=<redacted>
2026-07-07T11:25:27Z error [electron-message-handler] Received turn/started for unknown conversation conversationId=<redacted>
2026-07-07T11:27:30Z error [electron-message-handler] Received turn/completed for unknown conversation conversationId=<redacted>
2026-07-07T11:27:30Z error [electron-message-handler] Received turn/started for unknown conversation conversationId=<redacted>
2026-07-07T11:27:38Z error [electron-message-handler] Received turn/completed for unknown conversation conversationId=<redacted>

Screenshot matches the same fallback screen described in this issue: black background, red info icon, Oops, an error has occurred, and Update Codex / Try again buttons.

I am not attaching full logs publicly because they may contain account/session/user activity details. Sanitized snippets are included above.