Typing next message activates approval shortcuts

[grubbymediator]

What version of Codex is running?

codex-cli 0.65.0

What subscription do you have?

Business

Which model were you using?

gpt-5.1-codex-max

What platform is your computer?

Linux 6.17.7-200.fc42.x86_64 x86_64 unknown

What issue are you seeing?

When typing the next command while the agent is working, and an approval request pops up, it will be instantly approved (potentially forever) if the text contained letters "y" or "a".

› 1. Yes, proceed (y)
2. Yes, and don't ask again for this command (a)
3. No, and tell Codex what to do differently (esc)

What steps can reproduce the bug?

As in description

What is the expected behavior?

Maybe have approval shortcuts on harder-to-accidentally-trigger combinations or disable them completely.

Additional information

No response

[tosh]

I'm also running into this regularly

I think "No" is also triggered by "n" (?)

I like the proposal of making the shortcuts less easy to type (e.g. keyboard combination)

another way could be identifying that the user is typing a prompt (e.g. prompt field non-empty)

that doesn't fully solve the problem though because the user might also want to start typing a prompt at any time even if prompt field is currently empty.

[MNeMoNiCuZ]

I appreciate the hotkeys, always a fan of that, but when it interrupts another action, this is not good.
It's like a pop-up stealing focus while you are writing something. It's not a good experience at all!