Codex in VS Code: "Failed to release queued message send lock" and disappearing messages, how to fix
Back to the blog
Articlecodexopenaitroubleshootingvs codedev tools

Codex in VS Code: "Failed to release queued message send lock" and disappearing messages, how to fix

MafraOctober 6, 20265 min read

You press Enter, the input box clears and nothing happens. The message never shows up in the conversation, Codex does not answer, and sometimes it sits above the input with a spinner. If you open the log, you will see this:

[warning] Failed to release queued message send lock
errorMessage="\"undefined\" is not valid JSON"
errorName=SyntaxError

It is a bug in the Codex extension for VS Code (openai.chatgpt), not your network or your account. OpenAI shipped the fix in version 26.930.41038 on October 4, 2026. In most cases, updating the extension and reloading the window solves it.

Flowchart: check the Codex extension version, install the fixed version, reload the window and, if it still hangs, roll back to 26.917.62051
The three steps that fix most cases, and the way out if it still hangs.

What does "Failed to release queued message send lock" mean?

It means the extension could not release the lock that controls sending the message queue. With the lock stuck, your next message waits in the queue instead of reaching the agent.

The cause was described in issue #49834: the internal bridge between the extension and the UI serializes the response with JSON.stringify. When the lock release operation returns nothing, the result becomes undefined, and the JSON.parse on the other side fails with "undefined" is not valid JSON. An operation that succeeded is treated as an error.

Symptoms reported in issues #49988 and #50403:

  • Enter clears the input, but the message never appears in the conversation.
  • The message sits above the input with a spinner and a greyed-out trash button.
  • The message gets processed much later, up to about 1 hour, without you sending it again.
  • The same message runs twice, or Codex answers the previous request again.

There are no network, timeout or reconnect errors in the log. Switching chatgpt.followUpQueueMode between queue and steer, or changing chatgpt.composerEnterBehavior, did not help for the people who tried.

How do I confirm it is this bug?

Open the VS Code Output panel (View > Output), pick the Codex channel and look for the line Failed to release queued message send lock. If it shows up around the time your message vanished, this is your case.

Then check the installed version:

# macOS and Linux
code --list-extensions --show-versions | grep openai.chatgpt

# Windows (PowerShell)
code --list-extensions --show-versions | Select-String openai.chatgpt

Which extension versions have the bug?

The problem started with the October 1, 2026 update and was fixed on October 4, 2026. Release channel versions, checked on the VS Marketplace:

VersionPublishedStatus
26.917.620512026-09-23Before the bug. Used as rollback
26.928.314162026-10-01Has the bug
26.928.409062026-10-01Has the bug
26.930.215372026-10-02Has the bug
26.930.317302026-10-03Has the bug
26.930.410382026-10-04Fixed
26.930.612252026-10-06Latest release as of this date

Versions with 26.59xx in the number (for example 26.5928.31416) belong to the pre-release channel. Reports in issue #50403 show the bug there too.

How do I fix it?

Update the extension to the latest release and reload the window. --force makes VS Code swap the version even when the extension is already installed:

code --install-extension openai.chatgpt --force

If auto-update keeps you on an old version, ask for the fixed version explicitly:

code --install-extension openai.chatgpt@26.930.41038 --force

Then run Developer: Reload Window from the command palette (Ctrl+Shift+P or Cmd+Shift+P). The stuck message usually goes out on its own at that point, so check the conversation before sending it again, or it may run twice.

If you were on the pre-release channel, go back to release: in the Extensions view, open the Codex extension and click Switch to Release Version.

What if I use Remote-SSH or WSL?

The extension runs on the remote side. Run code --install-extension in the integrated terminal of the remote window, so it installs on the server or in WSL, not on your local machine. After the fix shipped, one user reported the problem persisted over Remote-SSH on Linux, while others on Remote-SSH and WSL confirmed it was solved.

What if it still hangs after updating?

Roll back to the last version before the bug and pin it, so auto-update does not bring the broken version back:

code --install-extension openai.chatgpt@26.917.62051 --force

From the UI, right-click the extension, choose Install Specific Version... and select 26.917.62051. Installed that way, VS Code keeps the extension pinned to that version. When you want to go back to normal, update again.

In that situation, reporting helps. The maintainer who closed issue #49988 asked anyone still seeing the problem to comment with: extension version, operating system, whether you use SSH or WSL, and whether Codex was idle or working when the message was sent.

Does reloading the window fix it by itself?

It fixes it right away, but it does not last. Developer: Reload Window releases the lock and sends what was stuck, but on the affected versions the problem comes back after a few messages. In issue #50403, the author also reported seeing a notice that the chat was open elsewhere after reloading. Use the reload to get unstuck now and the update so you do not need it again.

If you want a coding agent in VS Code without counting tokens, Verboo Code has its own extension and offers unlimited tokens.

Enjoyed this article?
Share knowledge with your network.
// Read also

Related articles