SkillRL: Evolving Agents via Recursive Skill-Augmented Reinforcement Learning
Paper
• 2602.08234 • Published
• 69
Introduction
SkillRL-Search-RL is a checkpoint for the Search environment. This model is part of the SkillRL framework, which leverages hierarchical reinforcement learning to solve complex, long-horizon control tasks by combining pre-trained skill latents with high-level policy optimization.
- Task: Search
- Stage: RL
Key Features
Experience-based Skill Distillation: Transforms successful trajectories into strategic patterns and failed ones into concise lessons from failure.
Hierarchical SKILLBANK: Organizes knowledge into General Skills for universal strategic guidance and Task-Specific Skills for category-level heuristics.
Recursive Skill Evolution: A dynamic mechanism where the skill library co-evolves with the agent's policy during RL by analyzing validation failures.
Context Efficiency: Achieves 10-20% token compression compared to raw trajectory storage while enhancing reasoning utility.
Download
You can download the model then run the training scipts in https://github.com/aiming-lab/SkillRL.
Citation
@article{xia2026skillrl,
title={SkillRL: Evolving Agents via Recursive Skill-Augmented Reinforcement Learning},
author={Xia, Peng and Chen, Jianwen and Wang, Hanyang and Liu, Jiaqi and Zeng, Kaide and Wang, Yu and Han, Siwei and Zhou, Yiyang and Zhao, Xujiang and Chen, Haifeng and others},
journal={arXiv preprint arXiv:2602.08234},
year={2026}
}