When the agent attempts to review/plan the next steps of an issue, it gets stuck in a loop of repeating the same message. It does eventually stop, but it never self-corrects or properly responds to my prompts. This has started since the new version was released yesterday. Third time this has happened.
Steps to Reproduce
This typically happens when I prompt the agent to do a task. It doesn’t follow the request, so I have to undo changes and provide a new prompt.
Yes, they are enabled, but it doesn’t necessarily happen when the model calls the MCP. Also, is more commun with the Gemini 2.5 Flash than 2.5 Pro. Soon as i got this error again i will send the request id of the chat.
@condor The issue persists in my tests. I just had this issue again yesterday in a different situation. Agent mode is set to auto, but I only have the models GPT-5 and GPT-5-High selected. I have zero MCPs configured or enabled.
@Justin_Clarke thank you for letting us know. Auto works independent of your model settings as we internally route to best model for the task and depending on model availability in case of outages.
If you could share Request IDs with privacy disabled it would allow us to look into the request at what causes the loop.
I’m having the same problem (auto mode). It responds automatically and changes some files, but at some point it gets stuck and keeps displaying the same message over and over again until I stop it. When I start a new chat, it behaves the same way again.
