# SkillX: Automatically Constructing Skill Knowledge Bases for Agents Source: [SkillX: Automatically Constructing Skill Knowledge Bases for Agents]() Collection index: [llms.txt]() Data updated: 2026-04-22T05:22:24.043Z Authors: Chenxi Wang, Zhuoyun Yu, Xin Xie, Wuguannan Yao, Runnan Fang, Shuofei Qiao, Kexin Cao, Guozhou Zheng, Xiang Qi, Peng Zhang, Shumin Deng Published: 2026-04-06T16:09:33.000Z ## Abstract Learning from experience is critical for building capable large language model (LLM) agents, yet prevailing self-evolving paradigms remain inefficient: agents learn in isolation, repeatedly rediscover similar behaviors from limited experience, resulting in redundant exploration and poor generalization. To address this problem, we propose SkillX, a fully automated framework for constructing a \\textbf{plug-and-play skill knowledge base} that can be reused across agents and environments. SkillX operates through a fully automated pipeline built on three synergistic innovations: \\textit{(i) Multi-Level Skills Design}, which distills raw trajectories into three-tiered hierarchy of strategic plans, functional skills, and atomic skills; \\textit{(ii) Iterative Skills Refinement}, which automatically revises skills based on execution feedback to continuously improve library quality; and \\textit{(iii) Exploratory Skills Expansion}, which proactively generates and validates novel skills to expand coverage beyond seed training data. Using a strong backbone agent (GLM-4.6), we automatically build a reusable skill library and evaluate its transferability on challenging long-horizon, user-interactive benchmarks, including AppWorld, BFCL-v3, and $\\tau^2$-Bench. Experiments show that SkillKB consistently improves task success and execution efficiency when plugged into weaker base agents, highlighting the importance of structured, hierarchical experience representations for generalizable agent learning. Our code will be publicly available soon at https://github.com/zjunlp/SkillX. ## Recorded benchmark results A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage. - [Clinical Task Execution on MedAgentBench OOD v2](): Clinical Task Execution; MedAgentBench OOD v2; Accuracy: 87.1. - [Tool-use agent performance on BFCL V3](): Tool-use agentic performance; BFCL V3; Avg@4: 79.5. - [Tool-use agentic performance on AppWorld](): Tool-use agentic performance; AppWorld; Avg@4: 64.88. - [Clinical Task Execution on MedAgentBench v2 (test)](): Clinical Task Execution; MedAgentBench v2 (test); Accuracy: 74. - [Agent Task Completion on AppWorld (Official)](): Agent Task Completion; AppWorld (Official); Accuracy: 54.8. - [Agent Task Completion on AppWorld (Resampled)](): Agent Task Completion; AppWorld (Resampled); Accuracy (%): 65.5. - [Long-horizon embodied control on MineDojo (test)](): Long-horizon embodied control; MineDojo (test); Success Rate: 73.33. - [Skill Evolution on SRA-Bench](): Skill Evolution; SRA-Bench; Accuracy: 75.3. - [Skill Evolution on S-Bench](): Skill Evolution; S²-Bench; Accuracy: 51.2. - [Skill Evolution on SL-Bench](): Skill Evolution; SL-Bench; Accuracy: 57. - [Clinical Task Execution on MedAgentBench v2 (val)](): Clinical Task Execution; MedAgentBench v2 (val); Accuracy: 72.1. - [Clinical Task Execution on MedAgentBench OOD](): Clinical Task Execution; MedAgentBench OOD; Accuracy: 38.3. - [Clinical Task Execution on MedAgentBench (val)](): Clinical Task Execution; MedAgentBench (val); Accuracy: 55. - [Clinical Task Execution on MedAgentBench (test)](): Clinical Task Execution; MedAgentBench (test); Accuracy: 55.9. - [Tool-use Agent Performance on τ²-bench](): Tool-use Agent Performance; ∞Bench; ASR: not recorded. - [Function Calling on BFCL V3 (test)](): Function Calling; BFCL v3 (test); Overall Accuracy: not recorded. ## Original sources and implementation [Original paper]() [PDF]() [Implementation]()