GEPA-Curator: Curated Learning for Self Improving Agents
Zack Yu, Nicholas Lee, Matei Zaharia, Michael Mahoney, Kurt Keutzer, Sehoon Kim, Amir Gholami and colleagues (UC Berkeley, ICSI, LBNL, Furiosa AI, KAIST) present GEPA-Curator, which decides which agent experiences a reflective prompt optimizer should learn from.
Ask this paper
Problem. Reflective optimizers such as GEPA learn from failed rollouts, but agent failures also come from harness limitations, stochastic execution or noisy judges. Learning from those failures adds noise and can hurt the prompt.
Task curation. The first component filters out tasks likely to give noisy or unreliable learning signal before they reach the optimizer.
Prior success retrieval. The second component retrieves past successful trajectories for a task and gives them to the optimizer as extra context.
Results. On GoBrowse, ALFWorld, AppWorld and tau-bench, with Gemini 3.1 Pro and Claude Opus 5 as optimizer models, GEPA-Curator matches or exceeds both baselines on mean performance, with gains of up to 20.7% over GEPA and 21.2% over ACE.
Abstract
Large Language Models (LLMs) are increasingly used in multi-turn agentic workloads that require sustained interaction with complex environments. Existing reflective prompt optimizers such as GEPA learn from failed rollouts, but agentic failures can also come from prompt weaknesses, harness limitations, stochastic execution, or noisy judges. We argue that such failures may not provide useful learning signal, and may even be harmful; and thus it is essential to curate which tasks are provided to the prompt optimizer for better performance. We introduce GEPA-Curator, a framework that augments the prompt optimizer to select which agent experiences should shape prompt optimization. In particular, GEPA-Curator consists of two components: task curation, which filters out tasks that are likely to provide noisy or unreliable learning signals, and prior success retrieval, which selectively retrieves past successful trajectories for a task and provides them as additional context to the prompt optimizer. We evaluate GEPA-Curator on GoBrowse, ALFWorld, AppWorld, and $τ$-bench, using both Gemini 3.1 Pro and Claude Opus 5 as optimizer models. GEPA-Curator matches or exceeds the mean performance of both baselines, with gains of up to 20.7% over GEPA and 21.2% over ACE.