happened to me too, it wanted to run an invalid command and no matter if i let it run and it fails or skip it, it just kept on running it over and over
This happened to me as well, in the GPT-5 Codex extension. It kept running inline Python programs to investigate the same things over and over again. It also broke things. It seemed drugged. Things were broken, but after I reverted all of its changes things worked fine - scary. I’ve used Claude ever since (~2 days). When I return to GPT-5 Codex, I will be SURE to set it on HIGH and nothing else from now on! This could be a Cursor interaction since I normally use GPT-5 Codex in the codex CLI and have never seen this kind of confused/lost looping behavior before. Maybe it was running below “Medium” level which the CLI uses by default?