Understanding and Exploiting Initialization Anchoring Weakness in Feedback-Based Agent Planning

T1Covered by 1 source · first reported by arXiv at 25 Sept, 04:00 UTC

arXiv:2609.29697v1 Announce Type: cross Abstract: Feedback-based planning improves agent reliability by incorporating tool observations and corrective feedback. However, its protection may not be distributed uniformly across planning stages. We conduct a round-wise analysis of four representative…

science

Sources

• T1Understanding and Exploiting Initialization Anchoring Weakness in Feedback-Based Agent PlanningarXiv25 Sept, 04:00