CLI bugs out whenever usage limit is hit or when it's handling multiple large files at once

[lukaszszafranski]

What version of Code is running?

code 0.2.80

Which model were you using?

GPT-5

What platform is your computer?

Windows

What steps can reproduce the bug?

Write few prompts to hit usage limits/Use /plan, /code commands on bigger task

What is the expected behavior?

Prompt should be able to proceed, continue with previous context and finish all tasks.

What do you see instead?

Whenever I ask CLI with commands that run multiple agents like /plan and /code, it's working only for two prompts maximum.

If the output is huge, it'll behave strangely - i.e, when I try to scroll down/up in CLI, it outputs "[B[A" characters in the chat, mouse click is "[I" character, etc. When I hit usage limit for any of the models i'm subscribed too, the same is happening. I'm unable to submit any further prompt or even leave the cli by using Ctrl+C. I need to close entire terminal, because otherwise, I don't have any control over it.

If that helps, i've used "Full-Auto" permissions inside of my settings.

I've CC, Gemini and Codex, I've all subs necessary for running this stack.

Additional information

No response

[lukaszszafranski]

Also it'd be great if we were able to see in CTRL+A multiple agents running -> not only via text, because I had gemini running for around an hour and I couldn't tell if it was running, stuck or doing anything at all.

[zemaj]

Could you explain this one a little more? "Whenever I ask CLI with commands that run multiple agents like /plan and /code, it's working only for two prompts maximum." do you mean one after the other?

Could you send over a screenshot of what you're seeing? We have had issues with character sanitisation in the past - perhaps we missed a spot.

[lukaszszafranski]

Hi @zemaj, if that helps in any way, I was alt-tabbing multiple times yesterday and did not try to just let it run with focus on terminal. I tried to recreate the issue today, but my claude and gemini agents are unable to start, because they try to run in my C:/Users/UserName/.code folder instead of running inside of my C:/Users/Username/Documents/Github/ProjectFolder/.code

[lukaszszafranski]

[zemaj]

Thanks for the detailed report (and the screenshots)! We tracked this down to a background panic that could happen when we hit rate limits or juggle multi-agent output; the UI would restore the terminal but keep the main loop alive, leaving you with the stray “[A/[B/[I” sequences and no working Ctrl+C.

This is fixed on main in 34ffe46 (will ship with the next release ≥ 0.2.154). In that build the panic hook always restores the terminal and then exits, so the shell is back in a clean state even if a worker thread crashes. Once you update, the CLI will simply exit instead of wedging. Please reopen if you can still reproduce it after upgrading.