You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Codex desktop repeatedly crashes when using Computer Use. #26458
Yes, this appears to be the same underlying crash signature as #26079: Codex Desktop 26.601.21317, triggered thread git, SIGABRT, node::OOMErrorHandler. My issue may add a reproduction clue: it occurs after invoking Computer Use, and my crash reports include browser-use-peer-authorization.node. I’m okay closing my issue as a duplicate if you want to track the core Node/V8 git-worker OOM in #26079, but I recommend noting the Computer Use/browser-use trigger there.
The Computer Use app does not currently feel reliable or stable enough, probably because it has to interact with macOS Gatekeeper and the hardened security model in a very complex way.
The main OpenAI developer responsible for the Codex Computer Use app on macOS needs to be highly experienced, deeply familiar with macOS internals, and extremely careful. A system that can potentially take full control of a user’s machine must be engineered with exceptional precision. To work properly with macOS’s strong security model, the implementation needs to integrate organically with Gatekeeper, code signing, notarization, TCC permissions, and hardened runtime behavior. In short, this kind of feature requires real mastery of macOS, not just ordinary desktop app development.
Dear OpenAI Dev Team:
And please, do not ship features before they have been thoroughly tested across a wide variety of real-world environments. Releasing something prematurely and losing user trust is far more damaging than delaying a feature for a little longer. What matters most right now is reliability and confidence in the product, not the rapid addition of minor features that many users may not even need immediately. It would be much better to prioritize stability, eliminate these kinds of bugs and issues, and ensure a consistently dependable experience before expanding functionality further.
I hit what looks like the same family of issues with the intermittent macOS Gatekeeper dialog:
"Codex Computer Use.app" is damaged and can't be opened. You should move it to the Trash. This file was downloaded on an unknown date.
In my case this did not look like an actually corrupted helper bundle.
Environment / observed state:
Apple Silicon macOS
Codex.app version observed after the latest update: 26.602.30954
As a temporary test, I created a symlink from the canonical path to the timestamped copy. That made the expected path valid, and codesign / spctl still passed.
After restarting Codex and going to Settings > Computer Use > Install, Codex replaced the symlink with a real app bundle directory at:
~/.codex/computer-use/Codex Computer Use.app
The timestamped copy remained. The new canonical bundle passed codesign and spctl, and the notify target under the canonical path existed.
After that, Computer Use worked successfully against Zed.app:
SkyComputerUseService was running.
SkyComputerUseClient was running.
get_app_state(app="Zed") succeeded.
The accessibility tree and screenshot were returned.
A click on Zed's File menu succeeded.
My interpretation:
This looks less like a bad download or broken notarization ticket, and more like a per-user Computer Use install-state / helper lifecycle / Gatekeeper interaction issue.
The issue may become visible intermittently because the relevant helper is not always launched at Codex startup. It may happen later when:
opening Settings > Computer Use
pressing Install
launching the per-user helper
launching a nested helper such as SkyComputerUseClient.app
running the turn-ended notify hook
attaching to the first target app
performing the first screenshot / accessibility operation
The most useful thing for maintainers may be to clarify and harden the expected lifecycle:
Treat ~/.codex/computer-use/Codex Computer Use.app as the canonical per-user install target, or document the intended target.
Ensure Codex startup and plugin refresh resync this target from the bundled app if it is missing or stale.
Avoid leaving config.toml notify hooks pointing to a missing canonical helper path.
Validate parent and nested helper apps consistently.
Clean up stale SkyComputerUseService, SkyComputerUseClient, node_repl, and stdio app-server helper processes after use.
Clean up inactive /private/var/folders/.../com.openai.codex.code_sign_clone/code_sign_clone.* directories after quit.
Avoid triggering repeated syspolicyd / trustd validation loops for helpers that are already signed and notarized.
This may explain why a reinstall alone did not help, but rerunning the Computer Use installer after the canonical per-user path was missing appeared to stabilize the system.
Activity
github-actions commentedon Jun 4, 2026
Potential duplicates detected. Please review them and close your issue if it is a duplicate.
Powered by Codex Action
Cloisterwood commentedon Jun 4, 2026
Yes, this appears to be the same underlying crash signature as #26079: Codex Desktop 26.601.21317, triggered thread git, SIGABRT, node::OOMErrorHandler. My issue may add a reproduction clue: it occurs after invoking Computer Use, and my crash reports include browser-use-peer-authorization.node. I’m okay closing my issue as a duplicate if you want to track the core Node/V8 git-worker OOM in #26079, but I recommend noting the Computer Use/browser-use trigger there.
pengusto commentedon Jun 4, 2026
I’m seeing a very similar issue after updating to Codex 26.602.30954.
Environment:
Error message:
The popup appears repeatedly, even when I’m not manually launching Computer Use. This started after the latest Codex update.
aifirstd3v commentedon Jun 5, 2026
Dear OpenAI Dev Team:
And please, do not ship features before they have been thoroughly tested across a wide variety of real-world environments. Releasing something prematurely and losing user trust is far more damaging than delaying a feature for a little longer. What matters most right now is reliability and confidence in the product, not the rapid addition of minor features that many users may not even need immediately. It would be much better to prioritize stability, eliminate these kinds of bugs and issues, and ensure a consistently dependable experience before expanding functionality further.
I hit what looks like the same family of issues with the intermittent macOS Gatekeeper dialog:
"Codex Computer Use.app" is damaged and can't be opened. You should move it to the Trash.This file was downloaded on an unknown date.In my case this did not look like an actually corrupted helper bundle.
Environment / observed state:
26.602.309543575809com.openai.sky.CUAServiceImportant paths observed:
Bundled source helper:
/Applications/Codex.app/Contents/Resources/plugins/openai-bundled/plugins/computer-use/Codex Computer Use.appPer-user Computer Use install target:
~/.codex/computer-use/Codex Computer Use.appPrevious / timestamped local copy:
~/.codex/computer-use/Codex Computer Use.app 13-51-16-627.appTemporary bundled marketplace path:
~/.codex/.tmp/bundled-marketplaces/openai-bundled/plugins/computer-use/Codex Computer Use.appRuntime code-sign clone paths:
/private/var/folders/.../X/com.openai.codex.code_sign_clone/code_sign_clone.*/Codex.app.bundle/.../Codex Computer Use.appWhat I verified:
The bundled source helper inside
/Applications/Codex.app/.../computer-use/passed:codesign --verify --deep --strictspctl --assess --type executeThe local timestamped helper copy also passed
codesignandspctl.The helper binaries had the same expected OpenAI Developer ID signature.
No quarantine xattr was observed.
SkyComputerUseServicehashes matched between the bundled source and the local copy.The local state mismatch I found:
~/.codex/config.tomlhad anotifyhook pointing to the canonical path:~/.codex/computer-use/Codex Computer Use.app/Contents/SharedSupport/SkyComputerUseClient.app/Contents/MacOS/SkyComputerUseClientBut at that point the canonical
~/.codex/computer-use/Codex Computer Use.appdid not exist. Only a timestamped copy existed:~/.codex/computer-use/Codex Computer Use.app 13-51-16-627.appAs a temporary test, I created a symlink from the canonical path to the timestamped copy. That made the expected path valid, and
codesign/spctlstill passed.After restarting Codex and going to Settings > Computer Use > Install, Codex replaced the symlink with a real app bundle directory at:
~/.codex/computer-use/Codex Computer Use.appThe timestamped copy remained. The new canonical bundle passed
codesignandspctl, and thenotifytarget under the canonical path existed.After that, Computer Use worked successfully against Zed.app:
SkyComputerUseServicewas running.SkyComputerUseClientwas running.get_app_state(app="Zed")succeeded.My interpretation:
This looks less like a bad download or broken notarization ticket, and more like a per-user Computer Use install-state / helper lifecycle / Gatekeeper interaction issue.
The issue may become visible intermittently because the relevant helper is not always launched at Codex startup. It may happen later when:
SkyComputerUseClient.appturn-endednotify hookcode_sign_cloneruntime copyThe most useful thing for maintainers may be to clarify and harden the expected lifecycle:
~/.codex/computer-use/Codex Computer Use.appas the canonical per-user install target, or document the intended target.config.tomlnotify hooks pointing to a missing canonical helper path.SkyComputerUseService,SkyComputerUseClient,node_repl, and stdio app-server helper processes after use./private/var/folders/.../com.openai.codex.code_sign_clone/code_sign_clone.*directories after quit.syspolicyd/trustdvalidation loops for helpers that are already signed and notarized.This may explain why a reinstall alone did not help, but rerunning the Computer Use installer after the canonical per-user path was missing appeared to stabilize the system.
cohaaan commentedon Jun 8, 2026
is this fixed?
Cloisterwood commentedon Jun 8, 2026
cohaaan commentedon Jun 9, 2026
mrgoonie commentedon Jun 13, 2026
workaround:
sudo xattr ~/.codex/computer-use/Codex\ Computer\ Use.appAhmet-Dedeler commentedon Jun 13, 2026
same issue
mrgoonie commentedon Jul 14, 2026
it happens again...