What it is
The repository’s own pitch: “ChatGPT thinks. Codex works.” ChatGPT handles planning while the Codex harness executes, so each system does what it is best at instead of one model doing both.
Why it showed up
5.7k stars within roughly a month. The interesting claim is structural rather than technical: splitting planning from execution across two different systems instead of waiting for one model to be good at both.
Who should look at it?
Developers comparing coding-agent workflows should read this as an orchestration experiment. The value is in the handoff: one system plans and another operates the repository through the existing harness.
What to inspect
Look for how context, approvals and tool results move between ChatGPT and Codex. A useful workflow needs a clear boundary for what is proposed, what is executed, and when a human can stop or revise the plan.
Watch point
Splitting planning and execution can make responsibilities clearer, but it also creates another integration boundary. Review what information is sent between systems and whether a stale plan can still trigger changes in the repository.