How is there not a /rewind or /revert feature?

[notnotjake]

Rewind or revert feature is very much needed. OpenCode and Claude Code have great support for undoing. Unless you want to commit on every change, not having this is insane and makes codex almost unusable.

[etraut-openai]

We previously exposed an experimental /undo command. It didn't get much use, and its design caused problems for many users, so we removed it. If you'd like to see something like this, please upvote this issue. We generally prioritize our feature development based on signal from users (as measured by upvotes).

I think that most Codex users don't see the need for /undo because they either use git to commit and revert changes or they rely on Codex to undo its own work when needed (e.g. "Undo that last change and redo using option 2").

[RobRoyce]

We previously exposed an experimental /undo command. It didn't get much use, and its design caused problems for many users, so we removed it. If you'd like to see something like this, please upvote this issue. We generally prioritize our feature development based on signal from users (as measured by upvotes).

I think that most Codex users don't see the need for /undo because they either use git to commit and revert changes or they rely on Codex to undo its own work when needed (e.g. "Undo that last change and redo using option 2").

This is a very naive point of view, doubly so now that you removed the ability to prohibit codex from editing files without permission / explicit approval.

Case in point, I asked Codex for a review of a document that was uncommitted. I wanted it to be solid before committing. Codex decides my design doc (that I spent an hour on) is "out of sync" with the current repo, so it deletes it and recreates a much shorter one. Nothing I could do to keep it from doing that (didn't ask for permission to edit the file because you removed that /permissions setting). Now I want to undo, but can't because you removed /undo. And guess what, that original file was 1000+ lines long. I asked Codex to undo its changes and it produced a doc that was ~450 lines long.

You can't remove safeguards and the ability to revert and have a model that can't reliably undo its own changes. This makes codex unusable for anyone doing serious work. Meanwhile Claude Code has safeguards and revert. How/why would you assume you shouldn't have a certain minimum of parity with your biggest competitor? Especially when it comes to safety and developer ergonomics?

[RobRoyce]

Lucky for me I have automatic backups using Time Machine, but this is honestly unacceptable.

[zba]

We previously exposed an experimental /undo command. It didn't get much use, and its design caused problems for many

You decided to revert alone ? why button still in extension (and not works) ?

this impress codex as buggy raw product. no consistency.

[LovesWorking]

We previously exposed an experimental /undo command. It didn't get much use, and its design caused problems for many users, so we removed it. If you'd like to see something like this, please upvote this issue. We generally prioritize our feature development based on signal from users (as measured by upvotes).

I think that most Codex users don't see the need for /undo because they either use git to commit and revert changes or they rely on Codex to undo its own work when needed (e.g. "Undo that last change and redo using option 2").

Hire better engineers? What a joke not to have a rewind action like Claude.

[chiyanking]

"restore conversation to this point" seems more like a context management operation, where the conversation state is reverted to an earlier message and everything after that point is ignored

[BorisMolch]

Can we distinguish between an "undo" feature and a "restore conversation to this point" feature?

My understanding is that a true "undo" might be more complex to implement reliably, since it would need to reverse prior actions and state changes. In contrast, "restore conversation to this point" seems more like a context management operation, where the conversation state is reverted to an earlier message and everything after that point is ignored.

I can implement the restore logic on my side, but I’m unsure about the correct approach to trimming the context. Specifically, how should I prevent messages after the restore point from being included in the next request? Should those messages simply be removed from the message history before sending the next prompt?

[emreertunc]

I personally didn't liked undo. Did not used it much but it wasn't feeling right. I don't know why.

I need exact /rewind feature as claude code cli. It asks me to rewind either conversation or code&conversation which is pretty useful when you are not ready to commit yet but want to try something.

With agentic coding I experiment with my codebase a lot and not being able to go back is pretty much blocks this process for me.

Also undo button exists on vscode extension right now but not working at all.

I hope we can get a proper rewind/revert/restore checkpoint feature which would make codex 2x useful for me.

[CLRT19]

/rewind let me get more control of my own context. The /rewind conversation feature is very important, cause you want to keep the context clean to avoid your codex from hallucinating when having too much irrelevant context.

In addition please also check out the /btw feature claude code added.

I don't agree that finding user don't use /undo is a sign that this feature is not needed, it could be that most users just don't know how to use it.

[CLRT19]

This feature helps with context management a lot.

[jogi47]

You can go back in the conversation at any point and see a fork icon on each of your messages, allowing you to create a local fork with context up to that point. (Codex native app)

But yes, this rewind will help us keep the context clean on same thread without this fork management.

Code undo is not priority people surely can wait for it but this rewind on context is really required. Also if they allow to create handoff like codex upto that point that would be awesome as well.

btw OpenAI is way behind on this appllication level AI Engineering.

[TaggartMaher]

Why /rewind is better than /undo or git commits

Basically, it's faster. It prevents confusion. We use this functionality far to frequently for git to be a viable option.

Git is for increments, /rewind is for LLM output experimentation

When I use the rewind feature of Claude Code, I usually am trying something out and I don't really have a solid base on how to do something yet I'm just probing. Usually a git branch would be the way to do this, just open a branch and mess around and use commits to revert if you break something.
The thing is, AI agents code so fast and this process has a cycle magnitudes higher frequency. So, I would end up creating hundreds or thousands of commits a day that I would not find useful.
I view AI chat interfaces like an expirament, I get "n" messages into the conversation, then I try n + 1, then I decide that taught me something at n - 3, so I want to walk back.

[navkast]

Hitting esc twice lets you change message history, which forks the conversation. This won't restore code, but it helps with context management.