Messages get stuck in send queue — "undefined" is not valid JSON on queued message lock release #49975

What version of the IDE extension are you using?

26.928.31416

What subscription do you have?

ChatGPT (account login, not API key)

Which IDE are you using?

VS Code

What platform is your computer?

Microsoft Windows NT 10.0.19045.0 x64

What issue are you seeing?

Messages sent in an ongoing conversation stop appearing in the chat (neither my sent message nor the model's reply ever show up), even though the extension log shows the Codex session otherwise running normally. The input box sometimes still shows the already-sent text after the fact, which looks like it never went through.

Extension log (Codex.log) shows this error at the moment it gets stuck:

[warning] Failed to release queued message send lock conversationId= errorMessage=""undefined" is not valid JSON" errorName=SyntaxError
at e.onFetchResponse (app-initial-.js)
at e.deliverMessage (app-initial-.js)
at e.deliverOrBufferMessage (app-initial-.js)
at e.handleMessage (app-initial-.js)
at windowMessageListener (app-initial-*.js)

Around the same time, repeated broadcasts are logged with no handler configured:
[warning] [IpcClient] Received broadcast but no handler is configured method=query-cache-invalidate
[warning] [IpcClient] Received broadcast but no handler is configured method=thread-queued-followups-changed

Steps to reproduce:

1. Start a conversation using gpt-6.1-sol, logged in via ChatGPT account (not API key).
2. Mid-conversation, change the reasoning effort using the model/effort selector (e.g. low -> medium -> high -> xhigh).
3. Send a new message. It never appears as sent in the thread and no reply arrives.
4. Reloading the VS Code window ("Developer: Reload Window") does NOT clear the stuck state — the same conversation/thread stays broken.

Workaround found: starting a brand new conversation works fine; the stuck state appears to be tied to the specific conversation/thread id, not the extension globally.

Separately, around the same timestamps I also see repeated filesystem errors that may be related:
[error] Request failed ... error={"code":-32603,"message":"O sistema não pode encontrar o caminho especificado. (os error 3)"} ... method=fs/getMetadata

CLI version installed separately: codex-cli 0.159.3 (updated via npm install -g @openai/codex).

What steps can reproduce the bug?

1. Open the Codex panel in VS Code and start a new conversation.
2. Select the gpt-6.1-sol model, logged in via ChatGPT account (not API key).
3. Exchange a few normal messages with reasoning effort set to "low" or "medium" — these work fine.
4. Mid-conversation, change the reasoning effort using the model/effort selector (e.g. low -> high -> xhigh).
5. Send a new message right after that change.

Expected: the message is sent and a reply is generated, same as before the effort change.

Actual: the message never appears as sent in the thread, no reply arrives, and the extension log records:
"Failed to release queued message send lock ... errorMessage="undefined" is not valid JSON errorName=SyntaxError at e.onFetchResponse (app-initial-*.js)"

6. Try "Developer: Reload Window" in VS Code — the same conversation/thread is still stuck afterward.
7. Open a brand new conversation — sending messages works normally again, confirming the stuck state is tied to that specific conversation/thread id.

What is the expected behavior?

Sending a message mid-conversation after changing the reasoning effort should work the same as before the change: the message appears as sent and the model replies, just like it did earlier in the same conversation at a lower effort level.

Additional information

Workaround: starting a brand new conversation resolves it immediately. The stuck state appears to be tied to that specific conversation/thread id, not to the extension globally — other conversations in the same VS Code window keep working fine.

Also noticed around the same timestamps, possibly related: repeated fs/getMetadata errors in the log ("O sistema não pode encontrar o caminho especificado" / os error 3), and "[IpcClient] Received broadcast but no handler is configured" warnings for thread-queued-followups-changed and query-cache-invalidate.

Separately: the Codex CLI itself (not the extension) currently refuses to use gpt-6.1-sol when logged in via ChatGPT account, returning:
{"type":"error","status":400,"error":{"type":"invalid_request_error","message":"The 'gpt-6.1-sol' model is not supported when using Codex with a ChatGPT account."}}
This may be a separate but related issue, since the model itself appears to have mixed support depending on login type.

Environment: Windows 10 (10.0.19045.0, x64), VS Code, extension openai.chatgpt v26.928.31416, Codex CLI v0.159.3 (installed separately via npm).