MEGA Hub

ActiveSaddler: Automated Curriculum Learning for Agent Harness Optimization

Authors

Do you know Sungho Park?You can claim authorship or link another user.Do you know Wonjoong Kim?You can claim authorship or link another user.Do you know Jue Zhang?You can claim authorship or link another user.Do you know Wook-Shin Han?You can claim authorship or link another user.Do you know Pengfei Gao?You can claim authorship or link another user.Do you know Chanyoung Park?You can claim authorship or link another user.Do you know Yongqiang Yao?You can claim authorship or link another user.Do you know Rao Fu?You can claim authorship or link another user.Do you know Elsie Nallipogu?You can claim authorship or link another user.Do you know Qingwei Lin?You can claim authorship or link another user.Do you know Victor Rühle?You can claim authorship or link another user.

Abstract

Automated harness optimization can substantially improve LLM agents by iteratively updating their prompts, tool interfaces, and control logic from execution feedback. However, existing methods primarily optimize how the harness is updated while largely fixing which training scenarios generate the feedback that drives those updates. As the harness evolves, the scenarios most useful for further optimization can change, suggesting that the training curriculum itself should adapt alongside the harness. We formulate this missing dimension of harness optimization as an automated curriculum learning problem and introduce ActiveSaddler. ActiveSaddler models the evolving curriculum as a non-stationary bandit with dynamically instantiated optimization targets. It abstracts recurring failures into reusable failure-pattern arms, estimates the potential learning progress from further targeting each pattern, and adaptively balances revisiting known weaknesses with exploring unseen scenarios for new ones. Optimization outcomes continually update both the set of discovered failure patterns and their priorities, allowing the curriculum to co-evolve with the harness. Experiments on GAIA2 and Terminal-Bench 2.0 show that ActiveSaddler consistently discovers stronger harnesses, improving test Pass@1 by 4.4 and 7.5 percentage points over the same harness optimizer using a scenario order fixed before optimization, respectively. Ablations further show that these gains depend on dynamically constructing optimization targets, estimating their evolving utility, and balancing continued optimization with new failure discovery. Together, these results establish automated curriculum learning as a new crucial optimization dimension for harness optimization.

Community

00