Skip to content

Conversation

@gapurov
Copy link

@gapurov gapurov commented Dec 4, 2025

This adds a new @react-grab/codex package that mirrors the existing Cursor/Claude integrations, but wired up to the Codex CLI instead.

The server and client follow the same shape and options as the other packages, just adapted for codex exec, and have been tested end‑to‑end to work with React Grab.

@vercel
Copy link

vercel bot commented Dec 4, 2025

@gapurov is attempting to deploy a commit to the Million Team on Vercel.

A member of the Team first needs to authorize it.

@aidenybai
Copy link
Owner

hey @gapurov, thanks for your contribution. could you add it in the packages/website/app/blog/agent/page.tsx as an example, and add it in .changeset/config.json?

@gapurov
Copy link
Author

gapurov commented Dec 5, 2025

@aidenybai, added necessary changes

@gapurov gapurov force-pushed the feat-react-grab-codex branch from c3bb3a6 to a8f35cc Compare December 5, 2025 16:05
@tt-a1i
Copy link

tt-a1i commented Dec 6, 2025

Really nice PR – the Codex CLI integration feels very clean and consistent with the existing providers. I had a couple of small questions while reading through it (apologies if I'm missing context here):

  1. For the default port: Codex uses \ (packages/react-grab-codex/src/constants.ts), which looks like the same port Opencode uses (packages/react-grab-opencode/src/constants.ts). In \ (packages/react-grab-codex/src/server.ts) it seems to just return if the port is already in use without logging anything. If a user happened to have another agent running on that port (or started things in the "wrong" order), would the Codex bridge effectively fail to start silently? Wondering if that's intentional, or if it might be worth either giving Codex its own port or at least logging a clear message when the port is taken.

  2. In the Codex README (packages/react-grab-codex/README.md), the getting started section suggests , but the package name is . From the outside I'd expect something like \ (or a pnpm equivalent) unless there's a separate binary published under the unscoped name that I'm not aware of.

Again, could totally be me reading this in isolation, but figured I'd surface these in case they're helpful. Overall the structure and naming look great – the Codex event normalization in particular is very thoughtful.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants