Approval mode disappeared from Codex IDE extension?

[atyro]

I use the Codex IDE extension in Cursor. This week the “approval mode selector” disappeared without any changes to the docs (mode selector is still there), or changelog (maybe it’s buried in there, but I can’t find it).

• Is anyone else seeing this?
• Is this a bug, or intentional?

I wish OpenAI would provide more visibility to developers. The extension is closed-source so there’s no way to investigate further, and the changelog is focused on the core CLI tool.

The mode selector is important: it provides message-scoped approvals and thus peace-of-mind (especially when queueing messages) that the agent will not make unauthorized changes while unsupervised.

Choose an approval mode - Codex IDE extension features

[EricGT]

Yes they did make some changes.

Have you seen this.

Best Practices for using Codex

As an avid user of VSCode Codex extension (Codex – OpenAI’s coding agent) and Plan mode with AI, now have both. Note: Plan mode with Codex, as opposed to the plan skill for Codex, is so new I do not find any official documentation at Codex Thanks to CSWYF3634076 · GitHub who noted this is now active. For vscode as the change is very recent, had to reinstall the extension to get to version 0.4.69 [Image] Then click “+” [Image] to reveal “Plan mode” selector [Image] When selected update…

[vb]

Yes, in the CLI /approvals is now /permissions.

I would expect the same for the IDE extensions.

[atyro]

@EricGT Thanks - I’d seen the “planning mode” changes, but I’m not sure why they would also have removed the “approval mode” toggle. I often queue up non-planning tasks (eg. “summarize the new classes created in the code you just wrote”) that require the model *not* make changes.

@vb This may have been moved to a slash command in the CLI, but in the extension the toggle is gone and that slash command is not available.

[EricGT]

@atyro

I’d seen the “planning mode” changes, but I’m not sure why they would also have removed the “approval mode” toggle.

I did not use the other way you noted.

In planning mode it can go for several rounds of you or the AI asking questions. Once the AI feels it has enough then it will present the plan asking if you approve, or have the option to give a follow-up reply which may or may not be interpreted as approval.

In short many users like the way other tools work and OpenAI seems to be moving towards that.

@atyro

I often queue up non-planning tasks (eg. “summarize the new classes created in the code you just wrote”) that require the model not make changes.

In working with AI and prompting, I can not remember a span of three months where what I was doing at the end of the three months was the same as when I started the three months.

It takes effort to keep up with all of the changes, I for one do not mind.

[atyro]

Agreed developers need to be flexible. Very fair trade-off for a super high pace of development!

In case any OpenAI folks are reading - I think the primary issue here is simply the need for clear communication on the IDE extension specifically (it’s already great for the CLI). Stale docs are fine but users shouldn’t be left wondering if the undocumented disappearance of a core feature is intentional or a bug.