Codex CLI 0.30 Only -  stream error: unexpected status 400 Bad Request: {"detail":"Instructions are not valid"};

same

Same here

<img width="762" height="436" alt="Image" src="https://github.com/user-attachments/assets/894201ca-9378-4559-aadf-f259db3bbbf3" />

Same on Mac OS after update by npm i -g @openai/codex
P.s. It's solved! 🎉🎉🎉

<img width="859" height="557" alt="Image" src="https://github.com/user-attachments/assets/0135b887-5c03-4e06-ada2-a7f63133e63b" />

Same on Mac OS after update by npm i -g @openai/codex

Seems we have a FUBAR build on both Windows and Mac... So at least we will properly get a quick fix...

same here

Yep, same issue here. They've probably broken the system instruction the backend API validates on requests.

Seems specific to 0.30.0, while a fix arrives (or update on OpenAI endpoints), workaround is going back to version 0.29.0:
npm install -g @openai/codex@0.29.0

same here

Likely broken by https://github.com/openai/codex/pull/3122 where they modified prompt.md which is strictly validated by the backend endpoints to validate Plus/Pro subscriber use through codex cli.

Tried to delete my Agent.md but did not resolve, so it is not our Instructions that are invalid

Same here.

same

+1 on linux

Seems OpenAI endpoits have been changed. It's fixed at 0.30.0.

It's solved. Everyone, try again

Hi folks - this should be fixed. Please let us know if you're still hitting errors!

Confirmed fixed for me.

I see this issue after updating it to CLI 0.36.0 on Linux

similar - codex-cli 0.41.0

tream error: unexpected status 400 Bad Request: {
  "error": {
    "message": "The encrypted content gAAA...o4o= could not be verified.",
    "type": "invalid_request_error",
    "param": null,
    "code": null
  }
}; retrying 1/5 in 194ms…

Yes, getting the same as @mgscox . This happens when I use openai api key for the requests.

unexpected status 400 Bad Request: {
  "error": {
    "message": "The encrypted content gAAA...3bk= could not be verified.",
    "type": "invalid_request_error",
    "param": null,
    "code": null
  }
}

This issue is closed and those are different 400 errors than this issue was reporting.

The cause on this issue was at 0.30.0 release the system prompt changed and the chatgpt.com/backend-api/codex/responses validates the Codex CLI system prompt - the mismatch was causing 400 rejection with {"detail":"Instructions are not valid"} message.

You guys are reporting 400 errors, but totally different error message - probably best to create a separate issue if you've not already done so.

If you have a plus subscription and re-auth using api-key (or another authentication) you cannot resume the conversation you must start a new one. I can only assume the prior auth-keys are in the headers somewhere which now conflicts with the new ones. I had this exact problem when I ran out of tokens and switched to API billing.

Resolvi o problema apagando a pasta .codex e reinstalando a extensão

> Same on Mac OS after update by npm i -g @openai/codex Seems we have a FUBAR build on both Windows and Mac... So at least we will properly get a quick fix...

Same
I've tried multiple times over the past month or two -- recreating API keys, reinstalling codex (brew install and with npm), watching it fail every time ... every damned time.

Never -not once - has codex ever run on my mac.

This is as best it gets:
unexpected status 400 Bad Request: {"success":false,"error":
{"name":"ZodError","message":"[\n {\n \"code\": \"invalid_union\",
\n \"errors\": [],\n \"note\": \"No matching discriminator\",\n
\"discriminator\": \"type\",\n \"path\": [\n \"tools\",\n 5,\n
\"type\"\n ],\n \"message\": \"Invalid input\"\n }\n]"}}