I guess we don’t know what the agent would do after seeing these warnings and a request for extra action.

Perhaps it would stop and rethink, perhaps it would focus on the fact that extra action is needed - and perform that automatically.

I suppose the decision would depend on multiple factors too (model, prompt, constraints).

Measure twice cut once seems to be forgotten these days.