Windows: npm install -g @openai/codex installs 0.120.0 but CLI fails with missing @openai/codex-win32-x64

Potential duplicates detected. Please review them and close your issue if it is a duplicate.

• #17423
• #17390

Powered by Codex Action

same problem

same problem

same problem

same problem

Same, I roll back to old version with Codex Desktop APP😅

same issue, roll back to 0.119.0 works

same problem

same problem

Same problem. Rolling back to version 0.119.0 can avoid this issue, temporarily​:
npm uninstall -g @openai/codex; npm install -g @openai/codex@0.119.0

same problem

We think this may have been caused by the fact that it took a while to upload all of the bundles to npm, and the win32-x64 bundle was the last one that was uploaded. If you hit this problem, try to uninstall and reinstall again now that all the bundles have been uploaded. Let us know if that fixes the problem.

Cleaning the cache of npm and re-installation work.

Cleaning the cache of npm and re-installation work.

It doesn't... tried all morning.

We think this may have been caused by the fact that it took a while to upload all of the bundles to npm, and the win32-x64 bundle was the last one that was uploaded. If you hit this problem, try to uninstall and reinstall again now that all the bundles have been uploaded. Let us know if that fixes the problem.

I also suspect the order by which bundles are uploaded to npm to be the issue as I just today had the same issue with codex 0.122.0.

I have a scheduler which checks if a new version is available (10 min interval) and if so builds a Docker image with it. Worked fine for several months now.

But with the image for codex 0.122.0 built today, I had this error:

file:///usr/local/lib/node_modules/@openai/codex/bin/codex.js:100
    throw new Error(
          ^

Error: Missing optional dependency @openai/codex-linux-x64. Reinstall Codex: npm install -g @openai/codex@latest
    at file:///usr/local/lib/node_modules/@openai/codex/bin/codex.js:100:11
    at ModuleJob.run (node:internal/modules/esm/module_job:437:25)
    at async node:internal/modules/esm/loader:639:26
    at async asyncRunEntryPointWithESMLoader (node:internal/modules/run_main:101:5)

Node.js v24.15.0

and then running npm install -g @openai/codex in the container fixes the issue.

So to me this looks like my scheduler found a new version of codex whgen that was uploaded to npm, but at that time the bundle for @openai/codex-linux-x64 was not yet uploaded to npm. Thus my image was built without @openai/codex-linux-x64.

The clean fix might just be to on-release upload the new codex bundle last and not before dependencies.

This commit would probably fix it: https://github.com/Wuodan/codex/commit/f1715ea946b3a637daaa64d9e906a64f6dc05b2b

It just makes sure that the tarball dist/npm/codex-npm-{version}.tg is published last.

PR is ready if someone invites it 😉

This is exactly the kind of case I had in mind when I made a small pre-check tool:

https://github.com/FelixWardUS/codex-npm-check

It checks the main Codex npm package and the selected platform packages before install, including whether their tarballs are reachable.

npm install -g codex-npm-check
cnc

For scripted upgrades or scheduled builds, this can help catch a half-published npm release before the install/image build grabs it. It only catches npm publishing/registry problems, not runtime bugs after install.

This will be fixed in the next release, thanks to this PR. Actually, it will be completely fixed one release after next because part of the fix is in the client code. The next CLI release will include the fix so it can avoid this problem on the following update.

The same problem was investigated for a long time