Codex CLI heading the wrong way mid-turn? On 0.163.0, a correction you send lands right away

#What changed

Codex CLI 0.163.0 shipped on Oct 15. The headline change is about steering: when you send a follow-up while Codex is still responding, it now preempts the response and takes effect right away. Before this, a mid-turn message generally had to wait for the current response to wrap up first.

The release also makes interrupts stronger. Stopping a turn now ends any active Code Mode cells and the nested tool calls they launched, so a long script Codex kicked off doesn't keep running after you've told it to stop. Values already stored in the session stay available.

Both changes were on by default in the code since Oct 7 (pull requests 51812 and 51835) and reached a stable release today. The same release also adds a searchable /config screen and lets Codex run independent read-only lookups, such as skills, memory, and conversation history, in parallel.

#Who is affected

Anyone using the Codex CLI in the terminal, on any plan, once they update to 0.163.0. The release notes are for the CLI only, so this doesn't tell you anything about the Steer and Queue behavior in the ChatGPT desktop app or on iOS.

#What to do now

• Update the CLI the same way you installed it, then run codex --version and check that it says 0.163.0.
• When you see Codex start down the wrong path, press Enter with a short correction instead of waiting for it to finish. The docs describe Enter while Codex is working as adding instructions to the current turn.
• If what you have is a next step rather than a correction, press Tab instead. That queues it for the next turn so it doesn't cut into the work in progress.
• Make the correction concrete: say what to stop and what to do instead, for example "don't touch the migrations, only change the API handler." A vague nudge that now lands instantly can derail a turn that was going fine.
• Use interrupt freely on long Code Mode runs or tool chains that are clearly off track. They now stop with the turn instead of finishing in the background.
• If you drive Codex from scripts or tests that assume a response always completes before new input is processed, re-check them after upgrading. OpenAI's own tests that depend on that order had to switch the new behavior off.

#What is not confirmed

The pull request notes that instant interrupts are still marked as a feature under development even though they're now on by default, so the behavior could still change. The release notes don't say whether the Codex IDE extension or desktop app pick up the same change. DotsFeed hasn't measured how much faster corrections land in practice.

#Sources

• Codex Releases on X: https://x.com/CodexReleases/status/2110847076934328501
• Codex CLI 0.163.0 release notes: https://github.com/openai/codex/releases/tag/rust-v0.163.0
• Enable instant interrupts by default (PR 51812): https://github.com/openai/codex/pull/51812
• Enable code mode interruption by default (PR 51835): https://github.com/openai/codex/pull/51835
• Codex CLI slash commands and shortcuts: https://developers.openai.com/codex/cli/slash-commands