We currently request the agent to start reAct loops from the agent query cell only, in append-only mode. What if the user pressed a button on a code cell, specified the changes it needed, and have the agent propose the code changes?
Is that not desirable due to the nature of notebooks? I get why, when sending a new reAct loop on the chatbar it should likely continue appending, but it'd be nice if it could edit a code or markdown on demand.