Low-quality thinking and nonsense action decisions were made with xhigh #24052

What version of Codex CLI is running?

codex-cli 0.133.0

What subscription do you have?

Pro Lite

Which model were you using?

gpt-5.5 (reasoning xhigh, summaries detailed)

What platform is your computer?

Linux 6.6.114.1-microsoft-standard-WSL2 x86_64 unknown

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

Windows Terminal

Codex doctor report

What issue are you seeing?

Allow me to quote a piece of thinking summary:

I’m thinking about whether the Kumo Input might not have a size property. Should I check the docs? It could be easier to use TypeScript for clarity. Maybe it’s best to avoid the size prop on Input altogether since there’s already a Select with a small size defined. If Input doesn't really need a size prop, I should probably just remove it for safety. I’ve noticed there’s no size usage for Input in the Kumo docs either.

Then it proceeds to remove the size prop from SOME field types of the new UI implemented in the same turn, only a few steps before. This makes the UI inconsistent since "there’s already a Select with a small size defined".

For the default experience without the detailed thinking summaries, the user may not even know why the model suddenly decides to remove a valid prop without actually seeing a lint or typecheck error, nor checking the docs, while it claims to have done so.

From what I can see, it's not unusual to spot some weird thinking summaries, which sometimes makes me wonder if the thinking model is the same as the one replying. But this is the first time I noticed that it is directly relatable to subsequent nonsense actions and didn't recover from the bad direction by itself.

What steps can reproduce the bug?

Uploaded thread: 019e4eb2-95f0-7143-87de-a7f930ceed68

What is the expected behavior?

Sensible thinking and decision-making

Additional information

No response