Auto-injected summarization and loss of context

I often use a “plan then update” approach where I have cursor generate a plan.md and then follow the plan to complete an update.

recently cursor has started regularly stopping in the middle of complex updates, summarizing the conversation history (often prioritizing info at the beginning) Reviewing a few updates that were done and then providing the summary and asking if there is anything else it can help with.

this leaves files in a bad state with partially completed updates and cursor seems lost in the sauce.. No idea what it was recently working on and no idea what files have been updated.

My only workaround is to open a new chat, provide the plan.md and git diff context. Ask cursor to review the plan and the diff and the current state of the codebase and hope it can find what still needs to be done.

This is a terrible experience and I want a way to disable this mid-conversation summarization and total loss of context.

Nothing in the current chat history is able to be located or “remembered” by the agent. It’s like all prior messages were removed from context and I have no way to get them back.


Happened again. I asked it to make a plan.md to address deviations from codebase standards / updates that ignored our .cursor\rules\gen-standards.mdc and README.md

It generated the plan.md and then immediately summarized and forgot all conversation history / all context removed from the conversation flow.

How can I prevent this auto-injected summarization?