Harnesses for Inference-Time Alignment over Execution Trajectories

ArXi:2605.21516v1 Announce Type: cross Harness engineering has emerged as an important inference-time technique for large language model (LLM) agents, aiming to improve long-term performance through task decomposition and guided execution. However, elaborate harnesses are not uniformly better: increasing decomposition or guidance can sometimes improve execution, but can also reduce final task success. We study harness design through the lens of inference-time trajectory alignment.