Current behavior of openai UX is that as soon as a response involves a call to a plugin end point, the openai UX shows the “regenerate” button… apparently indicating that the prompt processing is complete when in fact it is not.
One way to replicate with the todo example code, ask for detailed instructions on how to accomplish some task, then ask to have those items added to the todo list. This is a long running task but it almost immediately shows the 'regenerate" button which may confuse the user.
Suggest correctly detecting the end of the prompt processing via the usual means… the “processing” animation etc…