Bugged terminal - CODEX CLI (UI / UX)

Resolved 💬 5 comments Opened Jan 25, 2026 by YoloMark98 Closed Jan 26, 2026
💡 Likely answer: A maintainer (github-actions[bot], contributor) responded on this thread — see the highlighted reply below.

What version of Codex is running?

v0.91.0

What subscription do you have?

PRO

Which model were you using?

gpt 5.2 codex xhigh

What platform is your computer?

Ubuntu (Linux)

What terminal emulator and version are you using (if applicable)?

Terminal (Putty)

What issue are you seeing?

Hello, since I started using CODEX (several months ago now), there has been a significant terminal bug that can cause many users to make mistakes when operating the terminal. The bug in question is the “context window” remaining “left”, which is shown here:

<img width="615" height="150" alt="Image" src="https://github.com/user-attachments/assets/68aad5c0-2fe7-45b0-8ace-801ffbd44dd4" />

This context window always changes value while the agent is running, it is never reliable and I don't know why it hasn't been fixed for months now.

The second issue I encounter is the inability to scroll up the terminal when the agent is active. On other CLIs, this is possible, but on CODEX, it is still not possible, so most of the time, I am unable to check the modified code because it does not allow me to scroll up the terminal interface.

What steps can reproduce the bug?

Just when running the agents, for both issue

What is the expected behavior?

_No response_

Additional information

_No response_

View original on GitHub ↗

5 Comments

github-actions[bot] contributor · 7 months ago

Potential duplicates detected. Please review them and close your issue if it is a duplicate.

  • #9811
  • #7720
  • #9027
  • #9308

Powered by Codex Action

etraut-openai contributor · 7 months ago

Thanks for posting. Please open a separate bug report for each issue you want to report. This allows us to track, prioritize, and close issues independently.

Can you say more about the context window reporting? You've said that it's "always changing", which is expected behavior, since the agent is emitting messages and tools are being called, both of which consume context window space. You've also said that it's "never reliable". What behavior are you seeing that leads you to believe it's not reliable? Can you provide specific examples? Does this occur in specific sessions (e.g. with specific tool calls), or does it happen all the time?

We recently published a blog post about the Codex agent loop. It provides insights about how the context window works. You might find it informative.

For the second issue (relating to terminal scrolling), this may be specific to the terminal program you're using. I'm not familiar with Putty. Other terminal emulators allow you to scroll back into the history when Codex is running. You could try a different terminal program to see if that solves your problem.

wgtbgzas · 7 months ago

And What behavior are you seeing that leads you to believe it's reliable? Much times chatgpt and codex IS giving FALSE code. that consume plan and openai ROBBERY MONEY. for too simple script using latest chatgpt 5.2 or latest codex, it openai ROBBERY money. we have livestream, that for 30 min. it was giving false code until someone from livestream comment to CANCEL chatgpt and this what happens my co-worker just stop pay for openai ROBBERY MONEY. if someone is selling something make it work stable WE don't want buy beta. 68 euros month for something that DON'T work in robbery. and by the way something happens with this account that i write becouse this from openai just don't like it we are screenshots it, and maybe gonna fill complaints, chatgpt and codex is very UNRELIABLE just don't use it. Second BAD: yestarday we want to expose our PAT github credentials and chatgpt - codex refuse to do it. What command is US " we peoples " and not some codex - chatgpt to decide what to do. If i'm drinking water and im dying chatgpt - codex won't help me, NO. If i say, expose my github PAT only becouse codex web don't upload .bin on github, becouse he was BAD developerd. He has to do it. IT is against law. if i write sexual message to woman, chatgpt - codex has rights to stop me BUT I HAVE RIGHTS to stop money openai NOT robbery money, becouse codex - chatgpt decide NOT TO expose our github PAT. when I ROLD HIM expose them. very pissed. so peoples are violating rights, when they can't write sexual message to woman, BUT this iS NOT legal to expose when peoples SAY expose our github PAT password only to git push it. second IS AGAINST LAW. im paying for something that IS not illegal. BUT ILLEGAL IS openai to ROBBERY peoples money. where is github to put zero stars for openai.....

YoloMark98 · 7 months ago

<img width="771" height="510" alt="Image" src="https://github.com/user-attachments/assets/12a264c8-6fe0-4f0f-91fa-c4a52af0849e" />

for example here, i runned /status and i see 40% context window left, then compacted, because the context window left was 3-4%

etraut-openai contributor · 7 months ago

Sorry, but I'm very confused by your latest post. The screen shot you posted doesn't show any context window reporting.

You said "then compacted". Did you run the manual /compact command? Did you observe auto-compaction at this point? You didn't mention compaction in your original bug report.

I'm going to close this bug because 1) it's talking about multiple issues and 2) it's not clear what you're reporting. Feel free to open new bug reports.

If you see issues with compaction within a session, please use the /feedback slash command to upload your logs and session details and include the thread ID in the bug report.