OpenAI released Codex 0.159.0 with instant_interrupt, an optional feature that lets a new instruction reach Codex while it is responding or waiting on a long code operation.
OpenAI released Codex 0.159.0 with instant_interrupt, an optional feature that lets a new instruction reach Codex while it is responding or waiting on a long code operation. The feature is disabled by default.
If Codex starts working on the wrong file, for example, you can send a correction before its answer finishes. According to the model-response change, new input preempts the current wait and is passed into the continuing turn. That can correct the direction of the work sooner; it does not promise to undo changes already made.
Long exec and wait calls behave differently. The implementation notes say the call can yield a cell ID early so Codex can process the new instruction. The underlying command keeps running, and its output can be collected later. Sending a correction therefore does not by itself stop that command.
The same release lists Windows fixes intended to prevent stray console windows when starting MCP servers, code-mode hosts and piped commands. A user of an earlier CLI version reported such windows stealing focus while typing. That report shows why the fix matters, but it is not independent verification that every affected setup is fixed.