Please add a "pause to chat" without ending the subtask.

[devlux76]

I'm running into a problem where, if the AI picks up on the wrong thing to do in the subtask, I have to stop it which is fine. I talk to it, steer it a bit and resume. The task will eventually complete, but it has now been marked as canceled, so it doesn't return to the orchestrator.

I admit I might be doing this wrong, but the "principle of least surprise" means I'm doing it the way it looks like it's supposed to be done. Now, this isn't an issue if I don't have everything set to autoapprove, I can steer it at the next ask, but unchecking that, sort of defeats the purpose. When I'm working with this, I'm acting more as a senior dev/mentor and I like just watching it do its thing and then jumping in really quickly if I see it going off track. However, the fact it completes the subtask and doesn't return properly means I have to manually try to restart things with the orchestrator and that's sort of hit or miss.

[HadesArchitect]

Good point.