macOS Desktop Appshot fails: Unable to attach appshot, captureNotFound after start succeeds
This is a Codex Desktop macOS issue, not a Codex CLI issue.
Environment
- macOS: 26.5, build 25F71
- Codex Desktop: 26.527.31326, build 3390
- Computer Use helper:
com.openai.sky.CUAService, version 1.0, build 799
Symptom
When using Appshot from Codex Desktop, the screenshot appears to be taken, but Codex shows:
Unable to attach appshot
No appshot attachment is added to the conversation.
Reproduction
- Open Codex Desktop.
- Use Appshot / Attach app screenshot.
- Try with Finder or Zed as the target app.
- Codex shows
Unable to attach appshot.
The failure is not limited to one target app. It has reproduced with:
- Finder, bundle identifier
com.apple.finder - Zed, bundle identifier
dev.zed.Zed
Key Log Pattern
Latest observed log pattern:
Appshot hotkey pressed
target app found
capture request created
capture start request sent
capture worker received start
capture worker finished start with hasResponse=true
next-capture-update Apple Event failed
Codex Computer Use Apple Event error -10005: captureNotFound(requestID: "...")
capture settled failureReason=update_poll_failed hadAxText=false hadScreenshot=false status=failed
Concrete latest example:
2026-05-30T12:07:41.138Z requestId=8371604f-c1fb-491b-9dea-bbfb60bcb09e target Finder
2026-05-30T12:07:41.514Z capture start finished hasResponse=true
2026-05-30T12:07:42.026Z next-capture-update Apple Event failed captureNotFound
2026-05-30T12:07:42.027Z capture settled failureReason=update_poll_failed hadAxText=false hadScreenshot=false
Relevant logs are under:
~/Library/Logs/com.openai.codex/2026/05/30/
Especially:
codex-desktop-7f1787c9-5613-45e2-b949-61aef4129d42-17469-t0-i1-120154-0.log
codex-desktop-7f1787c9-5613-45e2-b949-61aef4129d42-17469-t2-i1-120212-0.log
Local Checks Already Performed
The following local causes were tested or ruled out:
- Codex was updated/reinstalled. Issue persisted.
- Full Disk Access was already granted to Codex.
- Disk write checks succeeded for:
~/.codex~/Library/Application Support/Codex$TMPDIR- The issue reproduces on Finder, so it is not target-app specific.
- Quarantine and stale helper duplicate cleanup were attempted.
- Computer Use helper rebuild/reset was attempted.
- Duplicate
SkyComputerUseServiceprocesses were killed.
After killing helper processes, Codex immediately spawned many new unmanaged helper instances again, and Appshot still failed with the same captureNotFound pattern.
Important Process Observation
Multiple SkyComputerUseService processes were observed, all parented by the Codex app process:
~/.codex/computer-use/Codex Computer Use.app/Contents/MacOS/SkyComputerUseService
launchctl showed them as unmanaged processes such as:
com.apple.xpc.launchd.unmanaged.SkyComputerUseS.<pid>
This suggests the helpers are spawned directly by Codex rather than managed by a normal LaunchAgent.
Current Diagnosis
This does not look like a macOS permission, disk write, reinstall, or target app problem.
The capture start request succeeds, but the subsequent capture update request fails because the Computer Use bridge cannot find the request ID it just created. The likely failure is inside the Codex Desktop Appshot bridge or Computer Use helper state/routing.
Working hypothesis:
The Appshot start request and update polling request are routed to different helper instances, or the helper loses its in-memory capture registry before Codex polls for the result.
This appears to be a product bug in Codex Desktop Appshot / Computer Use helper build 799 on macOS 26.5.
Expected Fix
Needs a Codex Desktop / Computer Use helper update that fixes Appshot capture request tracking.
Until then, the reliable workaround is to use a normal screenshot file attachment:
- Take a screenshot with macOS.
- Attach the image via
Add photos & files.
This avoids the Appshot bridge path that is currently failing.
9 Comments
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action
I reviewed the duplicate suggestions from the bot.
#23992 looks related, but that report appears to involve
SkyComputerUseServicecrashing / Apple Event-600. This issue has a different failure mode: capture start succeeds withhasResponse=true, then update polling fails with Apple Event-10005andcaptureNotFound(requestID: ...), and no screenshot or AX text is returned.#24775 also appears related, but this report includes the newer
captureNotFoundevidence and process observation on Codex Desktop 26.527.31326 / macOS 26.5.So this may be the same Appshot / Computer Use subsystem, but it is not exactly the same observable failure.
Update: still reproducible on the latest stable Codex Desktop available to me as of 2026-06-14.
Environment:
New reproduction:
Additional findings:
SkyComputerUseServiceprocesses for capture start and update polling. Pre-launching one helper does not change this behavior. This strongly supports the hypothesis that request state is stored in one helper process while polling is routed to another.-[NSWindow windowNumber], although this specific reproduction reachescaptureNotFoundrather than reporting a crash.This is therefore still present after several Desktop releases and does not appear to be a disk permission, TCC permission, stale registration, damaged install, or target-app issue. Please route this to the Codex Desktop / Computer Use owners; the likely fix is to reuse one service instance or make capture request state available to the polling instance.
Update: I performed a clean reinstall of Codex Desktop and Computer Use, without restoring my previous Codex state first. The issue still reproduces.
This rules out stale local state, old Computer Use helper files, cached app data, and permission drift as the root cause.
The previous local evidence still matches the current behavior: one Appshot attempt can spawn multiple SkyComputerUseService processes, and the helper logs socket lock contention:
socket lock is unavailable errno=35
So this appears to be a Codex Desktop / Computer Use multi-process IPC or request-state bug, not a user-side install or permission problem.
Additional reproduction on newer Codex Desktop build
I am seeing the same
captureNotFoundfailure mode on a newer Codex Desktop build.Environment
26.623.70822, build4559com.openai.codexcodex-cli 0.142.4codex-cli 0.142.0Darwin 24.6.0 arm64com.openai.sky.CUAService, version1.0, build857com.jetbrains.intellijUser-visible symptom
Every Appshot attempt fails with the toast:
Plain image attachment still works. The failing path appears specific to Codex Desktop Appshot / Computer Use capture, not general image upload.
Reproduction
DoubleCommand).Unable to attach appshotand no appshot is attached.Local repair attempts already tried
The issue still reproduces after:
SkyComputerUseService,SkyComputerUseClient, andbare-modifier-monitor.com.openai.codexcom.openai.sky.CUAServiceLog evidence
The latest attempts show the same pattern as this issue: start succeeds, but the update poll cannot find the capture request.
Earlier attempts from the same session show the capture start path completing before the failed update poll:
Diagnosis
This still looks like the Appshot start request and update polling request are not reaching the same capture registry/helper state. The helper reports that capture has started, but immediately afterward
next-capture-updatecannot find the request ID.That makes the issue look unlikely to be caused by:
Expected behavior
After the capture start request succeeds, the update polling path should find the same request ID and return screenshot/AX-text updates, or Codex should surface a more specific recovery/error state if the helper lost the request.
---
This issue or PR was generated by LazyCodex.
Tag: lazycodex-generated
I am seeing the same issue immediately after upgrading from the Codex desktop app to the ChatGPT desktop app today.
Environment
Reproduction
Unable to attach appshot.Checks already completed
I can provide a screenshot of the error toast if helpful.
I am getting this atm too
I’m seeing the same user-facing issue after the Codex -> ChatGPT desktop migration, but with a crash signature from the Computer Use helper.
Environment:
Apple Silicon
macOS 15.7.7 (24G720)
ChatGPT desktop app: 26.707.41301
Computer Use helper: 26.708.1000366
Behavior:
Crash logs:
SkyComputerUseService and SkyComputerUseClient crash at launch with:
DYLD Symbol missing: _swift_task_addPriorityEscalationHandler
Expected in: /usr/lib/swift/libswift_Concurrency.dylib
So in my case the Appshot helper appears to fail before any capture permission prompt or target-app attachment step.
Not sure if this will work for everyone, but the cause of my appshots breaking was I was symlinking ~/.codex some other directory. Replacing my symlinked ~/.codex with a real directory and clearing old Computer Use registrations fixed it. After restarting ChatGPT it worked again