Disable inline ghost suggestions in Codex CLI input

Hey folks, this is really destructing feature in terms of accessibility! i have to read all these suggestions every time whenever i read the line with my screen reader as a blind Codex-Cli user!
Could we please prioritize putting a convenient configuration to disable these inline suggestions on the message composer input?

This feature request hasn't received enough upvotes, so closing.

How does one upvote issues? The hints are at worst mildly annoying to me in a full terminal, but if I run in a limited one (like embedded emacs / neovim terminals, which have advantages because then agents can more easily communicate back to the editor) they become very distracting

This is probably similar to what @hasan-ozdemir is experiencing, although I imagine it's even more of a problem for someone using a screen reader.

Claude Code has a similar feature but made it controllable with an environment variable CLAUDE_CODE_ENABLE_PROMPT_SUGGESTION, which is sufficient for my purposes. I'd be pretty happy if codex had something comparable

how is this not planned? please consider it. it can't be hard to implement a config.toml user setting. It's very annoying to see that text in the prompt constantly.

The suggestions are distracting and unnecessary in my opinion.
As a workaround I overwrote the strings in the Codex binary with spaces using a hex editor:

"Explain this codebaseSummarize recent commitsImplement {feature}Find and fix a bug in @filenameWrite tests for @filenameImprove documentation in @filenameRun /review on my current changesUse /skills to list available skills"

But after a Codex update I have to do that again of course.
An option to disable the suggestions would be much more sensible. Please reconsider it.

Crazy that you won't give us the option. Is this a vibe code gimmick or a professional tool.

This feature request hasn't received enough upvotes, so closing.

juxtaposition with comment above really gets across "accessibility only impacts a fraction of users, so closing."

I'll use OpenCode instead. I can't imagine why OpenAI would ignore this, but anyway, good job.

EDIT: well, OpenCode isn't so great with GPT models... so I built my own codex with all the prompts removed.

This feature request hasn't received enough upvotes, so closing.

Not enough votes! What the heck? This is causing me issues too. I have a multi agent setup. They read each other's screens,and this suggested prompt thing is really throwing them off. Not to mention, there's like... two of them?

Codex has a serious configuration issue. So many things are just turned on because... well, I don't know. And they can't be turned off?

> This feature request hasn't received enough upvotes, so closing. Not enough votes! What the heck? This is causing me issues too. I have a multi agent setup. They read each other's screens,and this suggested prompt thing is really throwing them off. Not to mention, there's like... two of them? Codex has a serious configuration issue. So many things are just turned on because... well, I don't know. And they can't be turned off?

i highly recommend pi.dev (https://pi.dev/) i've been running this for several weeks now, much better experience.

The gray text is the composer's placeholder, not output from a suggestion engine — which is why nothing relevant shows up in codex features list and why [features] suggestions_feature = false has no effect.

At startup ChatWidget samples one of eight example prompts and renders it dim in the empty composer:

// codex-rs/tui/src/chatwidget/constructor.rs
let mut rng = rand::rng();
let placeholder = PLACEHOLDERS[rng.random_range(0..PLACEHOLDERS.len())].to_string();

Because that text sits exactly where typed input goes, it reads as pre-filled content rather than a hint, and the sampled example is unrelated to whatever you actually opened Codex to do.

I put together a patch that adds a [tui] setting, alongside the existing animations / show_tooltips preferences:

[tui]
show_prompt_suggestions = false

With it off, the composer shows the neutral Ask Codex to do anything label instead. The default stays true, so current behaviour is unchanged.

• Branch: https://github.com/mkavinkumar1/codex/tree/fix/tui-disable-prompt-suggestions
• Diff: https://github.com/openai/codex/compare/main...mkavinkumar1:codex:fix/tui-disable-prompt-suggestions

It also moves the two placeholder tables out of chatwidget.rs into a new chatwidget/placeholders.rs, since AGENTS.md calls out that file as one to keep to orchestration.

Tested on the pinned 1.95.0 toolchain: codex-tui --lib (3328 pass), codex-core --lib config:: (459 pass), cargo check -p codex-thread-manager-sample, and config.schema.json regenerated via just write-config-schema (byte-identical). Adds two unit tests plus an insta snapshot for the suggestion state.

I can't open a PR — this repo appears not to accept them from forks — and I see external contributions are by invitation only, so I'm leaving the branch here in case it's useful, or in case a maintainer would like to take it.

Just gonna chime in here and say this is highly annoying and a UX anti-pattern. It isn't even grayed out on my screen, and doesn't let me as a user know where I should type.

<img width="1218" height="192" alt="Image" src="https://github.com/user-attachments/assets/b54ef870-ff0e-46ec-9146-1f591307636c" />

<img width="1218" height="192" alt="Image" src="https://github.com/user-attachments/assets/9cf0bc36-634a-4aa1-ac2f-8c52e4744200" />

The above two screenshots do not give me any indication that I can type there. Horrible UX.

It should:

• be greyed out
• be disableable

Based on feedback, we've decided to remove the rotating placeholders, which users found confusing. We've replaced it with a generic placeholder message.

• https://github.com/openai/codex/issues/13466#issuecomment-5211502344