Sometimes I see the agent starting to write things that do not make sense and I want to intervene midway to put it on the right path, but without reverting completely to the previous stage. I imagine this is an LLM limitation rather than cursor’s, but I thought I’ll try and see if there’s a workaround.
Its certainly possible, sometimes you need to Skip commands in order to activate it, the time in which it asks you how to proceed isn’t perfect but good to intervene mid-work, there are two topics with custom tools including mine which requires some technical skill to make it work but I’ll create an easier solution in future:
security note: to stay safe you need to review code when it comes from custom tools not reviewed by vscode or glama/smithery