HGM-Dual: Lineage-Aware Guided Agent Harness Optimization for Long-Horizon Planning
HGM-Dual reports matching HGM variants’ final planning performance with 2–4 times fewer evaluations.
The framework improves an agent’s prompts, tools and workflows while leaving the underlying language model frozen. It records which changes helped or hurt across successive agent versions, giving later edits a traceable history. It then generates both exploratory variants and targeted variants aimed at the parent agent’s highest-ranked constraint violations. The authors report up to a 14.45% relative improvement in case accuracy over the strongest baseline.
ArXiv · AI/CL/LG's note
The framework improves an agent’s prompts, tools and workflows while leaving the underlying language model frozen. It records which changes helped or hurt across successive agent versions, giving later edits a traceable history. It then generates both exploratory variants and targeted variants aimed at the parent agent’s highest-ranked constraint violations. The authors report up to a 14.45% relative improvement in case accuracy over the strongest baseline.
ArXiv · AI/CL/LG's note
score 5