arXiv AI

Skill or Skip? Learning Selective Skill Invocation in Agentic Tasks via Dual-Granularity Preference Learning

arXiv:2606. 00510v1 Announce Type: cross Abstract: Agent skills are callable procedural modules that provide reusable knowledge and execution policies for complex agentic tasks.

arXiv AI
Oct 2

SkillLens: Adaptive Multi-Granularity Skill Reuse for Cost-Efficient LLM Agents

SkillLens introduces a hierarchical skill-evolution framework that organizes skills into a four-layer graph of policies, strategies, procedures, and primitives, allowing retrieval at mixed granularity. The system first retrieves semantically relevant skill seeds, expands them via a degree‑corrected random walk, and uses a verifier to decide whether to accept, decompose, rewrite, or skip each visited unit. This approach enables agents to reuse compatible subskills while locally adapting mismatched components, and theoretical analysis shows sublinear cost under sparse mismatch assumptions, with empirical results on MuLocbench and ALFWorld demonstrating consistent improvements over strong baselines.

By Ziyang Yu, Yongliang Miao, Liang Zhao, Bowen Zhu, Hasibul Haque
arXiv AI
Jun 2

ReSkill: Reconciling Skill Creation with Policy Optimization in Agentic RL

arXiv:2606. 01619v1 Announce Type: new Abstract: Agentic reinforcement learning (RL) enables LLM agents to improve continuously from environment rewards, yet the resulting policies do not systematically accumulate reusable strategies that generalize across tasks.

By Zelin He, Haotian Lin, Boran Han, Wei Zhu, Haoyang Fang, Bernie Wang, Xuan Zhu, Runze Li, Matthew Reimherr
arXiv AI
Jul 14

AgentAbstain: Do LLM Agents Know When Not to Act?

arXiv:2607. 10059v1 Announce Type: new Abstract: Agent systems based on large language models (LLMs) are increasingly deployed for autonomous tasks, yet existing evaluations mostly focus on task success rather than whether agents know when to abstain.

By Xun Liu, Yi Evie Zhang, Vira Kasprova, Parisa Rabbani, Pardis Sadat Zahraei, Tianyu Zhang, Ali Ebrahimpour-Boroojeny, Varun Chandrasekaran
arXiv Machine Learning
Oct 23

MOER: Mix-Policy Optimization under Expert Prefix Guidance with Selective Rollback

MOER is a mix‑policy optimization framework that blends expert guidance with self‑exploration in reinforcement learning. It injects a decaying expert chain‑of‑thought prefix into prompts, progressively annealing the prefix length to encourage exploration, and uses a selective rollback mechanism to replace failed completions with the full expert trace, ensuring correctness. The method employs a harmonized mix‑policy objective that optimizes on‑policy tokens with a GRPO loss while supervising both expert prefix and completion tokens with a trust‑region SFT loss, leading to stable training and superior performance on mathematical and general reasoning tasks.

By Xuyang Zhong, Jiuding Yang, Yiwen Guo, Chen Liu
arXiv AI
2d ago

Skill Adapter: Personalized Modification and Online Routing of General-Purpose Skills

Skill Adapter is a system that personalizes and routes general‑purpose skills for large language model agents. It learns local adjustments from user requirements and execution feedback, storing them as Adapters linked to each skill, and a routing agent selects appropriate skills and Adapters based on the current subtask and environment. The approach improves task success rates and reduces execution time on ALFWorld, WebShop, and SkillsBench, maintaining performance even as the skill pool scales to 1,000 skills.

By Qijia Zhuang, Yanyi Gou, Wenbin Ma, Yuxiang Ren
arXiv AI
Jun 6

LatentSkill: From In-Context Textual Skills to In-Weight Latent Skills for LLM Agents

arXiv:2606. 06087v1 Announce Type: cross Abstract: Agent systems increasingly use textual skills to encode reusable task procedures, but injecting these skills into the prompt at every step incurs substantial context overhead and exposes skill content as plaintext.

By Aofan Yu, Chenyu Zhou, Tianyi Xu, Zihan Guo, Rong Shan, Zhihui Fu, Jun Wang, Weiwen Liu, Yong Yu, Weinan Zhang, Jianghao Lin
arXiv AI
Sep 28

CODESKILL: Learning Self-Evolving Skills for Coding Agents

CODESKILL is an LLM-based framework that learns to extract, evolve, and maintain procedural skills from coding-agent trajectories. It treats skill extraction and skill-bank management as a learnable policy trained with reinforcement learning, using a hybrid reward combining rubric-based skill quality and verifiable execution feedback. Experiments on EnvBench, SWE-Bench Verified, and Terminal-Bench 2 demonstrate that CODESKILL raises average pass rates by 11.03 over a no-skill baseline and by 5.10 over the strongest prompt-based or memory baseline while keeping a compact skill bank.

By Yanzhou Li, Yiran Zhang, Xiaoyu Zhang, Xiaoxia Liu, Yang Liu
arXiv AI
Aug 24

AUSO: Action-Level Unified Skill Optimization from Internalization to Utilization

AUSO (Action-level Unified Skill Optimization) is a method that unifies skill learning and skill use through a progressive, action-aware optimization process. It starts by jointly learning from teacher guidance and environmental outcomes, then shifts to outcome-based policy optimization, and finally evaluates each action under skill-conditioned and skill-free contexts to strengthen beneficial skill-sensitive actions while suppressing harmful ones. Experiments on ALFWorld, WebShop, and SearchQA demonstrate that AUSO consistently improves agent performance and out-of-distribution generalization compared to competitive baselines.

By Huizu Lin, Chengkai Huang, Tianqi Gao, Tao Huang, Daijiao Liu, Tongxin Li, Xiaoyan Sun, Lina Yao
arXiv Machine Learning
Sep 22

Beyond Task Completion: Training Capable and Safe Computer-Use Agents

The paper introduces SCOPE, a method that post‑trains computer‑use agents to balance task completion with safety by conditioning actions on environmental risk. It combines supervised fine‑tuning on three trajectory types—capability demonstrations, safe continuations, and explicit refusals—followed by reinforcement learning to improve performance. Experiments starting from Qwen3.5‑9B show that SCOPE‑RL achieves high task success and attack‑avoidance rates, outperforming other agents on OSWorld and OS‑BLIND benchmarks.

By Zeyu Kang, Zhenyun Yin, Yang Zhang, Shan He, Shanzhe Lei, Yanjiu Zhong, Xinquan Chen, Yuhong Wang