Hi @danperks, I understand there can be limitations from 3rd party providers and/or cursor’s API. It makes sense.
That being said, this keeps interrupting the composer in agent mode. Since I can’t precisely track and restart it where it was interrupted, I gave to restart the agent from the beginning of the task, very likely to hit the limit again.
A very light retry mechanism with exponential back off seems to me like a good way to handle this issue with minimal effort. What do you think?