The paper introduces GEPA-Curator, an add-on for reflective prompt optimizers that decides which agent experiences should shape the prompt.
Task curation: Filters out tasks likely to give noisy or unreliable learning signals, since agent failures can come from prompt weaknesses, harness limits, stochastic execution or noisy judges.
Prior success retrieval: Pulls past successful trajectories for a task and gives them to the prompt optimizer as extra context.
Results: Tested on GoBrowse, ALFWorld, AppWorld and τ-bench, with Gemini 3.1 Pro and Claude Opus 5 as optimizers. The authors say it matches or exceeds the mean performance of GEPA and ACE, with gains of up to 20.7% over GEPA and 21.2% over ACE.
Sign in to comment.
TL;DR of the paper, generated automatically.
The paper introduces GEPA-Curator, an add-on for reflective prompt optimizers that decides which agent experiences should shape the prompt.