CHART: A Harness-Rotation Curriculum for Harness-Robust Search Agents

Search agents are usually trained under a single harness. But once an agent is deployed in a real application, its harness is frequently updated (e.g., a rewritten system prompt) to fit production needs. This exposes a fragility of post-trained agents: because a learned behavior is entangled with its training harness, even a harness update that leaves the task unchanged can fail to elicit the behavior. In this study, we train a search agent to perform parallel search, a popular strategy for improving both the efficiency and the performance of search. We find that training under a fixed harness makes the behavior \emph{harness-local}, overfit to that harness's surface form: when the harness changes, the model falls back to serial search. An intuitive idea is \emph{harness augmentation}, but simply creating more harnesses for training does not completely resolve the problem. GRPO learns from the reward gap between parallel and serial rollouts of the same question: a small harness pool saturates that gap early, while a large pool dilutes the per-harness signal too thinly for any harness to consolidate. We therefore propose Curriculum HArness Rotation Training (\textbf{CHART}), a rotating curriculum that lets a search agent gradually consolidate parallel search across harnesses. At each periodic evaluation, CHART ``graduates'' the harnesses on which the expected behavior has been learned and replaces them with harnesses that are still learnable, which keeps the reward gap alive throughout training. Starting from the same harness pool, CHART makes the model learn parallel search on all harnesses, whereas static augmentation succeeds on at most half of them. The behavior also carries to held-out harnesses: CHART parallelizes on $89\%$ of held-out turns, against at most $5\%$ for the static pools. It further transfers to a new QA task and search environment, improving \passat{1} by $5.6$pp over the best static pool. Besides, we show that CHART-trained agents benefit more from meta-harness search than other baselines.