[{"data":1,"prerenderedAt":4},["ShallowReactive",2],{"readme:skillx":3},"\u003Cdiv align=\"center\">\n\u003Ch1> 👉 SkillX 👈 \u003C\u002Fh1>\n\u003Cb>SkillX: Automatically Constructing Skill Knowledge Bases for Agents\u003C\u002Fb>\u003Cp>\u003Ca href=\"https:\u002F\u002Fgithub.com\u002Fzjunlp\u002FSKillX\" rel=\"nofollow ugc noopener\">\u003Cimg src=\"https:\u002F\u002Fawesome.re\u002Fbadge.svg\" alt=\"Awesome\" \u002F>\u003C\u002Fa> \n\u003Ca href=\"https:\u002F\u002Fopensource.org\u002Flicenses\u002FMIT\" rel=\"nofollow ugc noopener\">\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fbadge\u002FLicense-MIT-green.svg\" alt=\"License: MIT\" \u002F>\u003C\u002Fa>\n\u003Cimg src=\"https:\u002F\u002Fimg.shields.io\u002Fgithub\u002Flast-commit\u002Fzjunlp\u002FSKillX?color=green\" alt=\"\" \u002F> \u003C\u002Fp>\n\u003C\u002Fdiv>\u003Ch2>Table of Contents\u003C\u002Fh2>\n\u003Cul>\n\u003Cli>👀\u003Ca href=\"#overview\" rel=\"nofollow ugc noopener\">Overview\u003C\u002Fa>\u003C\u002Fli>\n\u003Cli>🔧\u003Ca href=\"#installation\" rel=\"nofollow ugc noopener\">Installation\u003C\u002Fa>\u003C\u002Fli>\n\u003Cli>🏃\u003Ca href=\"#quick-start\" rel=\"nofollow ugc noopener\">Quick Start\u003C\u002Fa>\u003C\u002Fli>\n\u003Cli>🎁\u003Ca href=\"#acknowledgement\" rel=\"nofollow ugc noopener\">Acknowledgement\u003C\u002Fa>\u003C\u002Fli>\n\u003Cli>🚩\u003Ca href=\"#citation\" rel=\"nofollow ugc noopener\">Citation\u003C\u002Fa>\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch2>📖 Overview\u003C\u002Fh2>\n\u003Cp>\u003Cstrong>SkillX\u003C\u002Fstrong> is a fully automated framework that constructs a \u003Cstrong>reusable, plug-and-play skill knowledge base\u003C\u002Fstrong> for LLM agents from experience.\u003C\u002Fp>\n\u003Cp>Instead of storing raw trajectories, workflows, or loosely structured reflections, SkillX distills agent experience into a \u003Cstrong>three-level skill hierarchy\u003C\u002Fstrong>:\u003C\u002Fp>\n\u003Cul>\n\u003Cli>\u003Cstrong>Planning Skills\u003C\u002Fstrong> for high-level task organization\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Functional Skills\u003C\u002Fstrong> for reusable tool-based subroutines\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Atomic Skills\u003C\u002Fstrong> for execution-oriented tool usage patterns\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>Built with a strong backbone agent, SkillX produces a transferable skill library that can be directly plugged into weaker base agents and new environments. Across challenging long-horizon, user-interactive benchmarks such as \u003Cstrong>AppWorld\u003C\u002Fstrong>, \u003Cstrong>BFCL-v3\u003C\u002Fstrong>, and \u003Cstrong>τ2-Bench\u003C\u002Fstrong>, SkillX consistently improves both \u003Cstrong>task success\u003C\u002Fstrong> and \u003Cstrong>execution efficiency\u003C\u002Fstrong>.\u003C\u002Fp>\n\u003Cdiv align=\"center\">\n  \u003Cimg src=\"https:\u002F\u002Fraw.githubusercontent.com\u002Fzjunlp\u002Fskillx\u002FHEAD\u002Fassets\u002Foverview.png\" alt=\"Case GIF\" \u002F>\n\u003C\u002Fdiv>\u003Chr \u002F>\n\u003Ch2>Data Formats\u003C\u002Fh2>\n\u003Ch3>Trajectory Input (JSONL)\u003C\u002Fh3>\n\u003Cp>SkillX expects trajectories in the following schema:\u003C\u002Fp>\n\u003Cpre>\u003Ccode class=\"language-json\">{\n  \"trajectory_id\": \"traj_001\",\n  \"task_id\": \"task_001\",\n  \"user_task\": \"How many songs are in my Spotify library?\",\n  \"task_history\": [\n    {\"role\": \"system\", \"content\": \"You are a helpful assistant...\"},\n    {\"role\": \"assistant\", \"content\": \"I'll help you count...\"},\n    {\"role\": \"user\", \"content\": \"Output:\\n```\\n{\\\"songs\\\": 150}\\n```\"}\n  ],\n  \"reward\": 1.0,\n  \"metadata\": {}\n}\n\u003C\u002Fcode>\u003C\u002Fpre>\n\u003Ch2>🤖 Key Features\u003C\u002Fh2>\n\u003Ch3>Hierarchical Multi-Level Skill Design\u003C\u002Fh3>\n\u003Cp>SkillX transforms raw trajectories into a structured three-tier skill space:\u003C\u002Fp>\n\u003Cul>\n\u003Cli>\u003Cstrong>Planning Skills\u003C\u002Fstrong> capture high-level decomposition and ordering\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Functional Skills\u003C\u002Fstrong> represent reusable multi-step tool subroutines\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Atomic Skills\u003C\u002Fstrong> encode practical tool usage constraints and patterns\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch3>Fully Automated Skill KB Construction\u003C\u002Fh3>\n\u003Cp>SkillX provides an end-to-end automated pipeline that:\u003C\u002Fp>\n\u003Cul>\n\u003Cli>rolls out agents on training tasks,\u003C\u002Fli>\n\u003Cli>extracts reusable skills from successful trajectories,\u003C\u002Fli>\n\u003Cli>consolidates and filters low-quality skills,\u003C\u002Fli>\n\u003Cli>and builds a reusable \u003Cstrong>plug-and-play skill knowledge base\u003C\u002Fstrong>.\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch3>Iterative Skill Refinement\u003C\u002Fh3>\n\u003Cp>SkillX continuously improves the skill library through:\u003C\u002Fp>\n\u003Cul>\n\u003Cli>\u003Cstrong>skill merging\u003C\u002Fstrong> for consolidating redundant behaviors,\u003C\u002Fli>\n\u003Cli>\u003Cstrong>quality filtering\u003C\u002Fstrong> for removing brittle or hallucinated skills,\u003C\u002Fli>\n\u003Cli>and \u003Cstrong>iterative updates\u003C\u002Fstrong> that add, modify, or keep skills based on execution feedback.\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch3>Exploratory Skill Expansion\u003C\u002Fh3>\n\u003Cp>Beyond seed demonstrations, SkillX proactively discovers new skills by:\u003C\u002Fp>\n\u003Cul>\n\u003Cli>identifying under-used and failure-prone tools,\u003C\u002Fli>\n\u003Cli>guiding environment exploration,\u003C\u002Fli>\n\u003Cli>synthesizing new tasks from exploratory trajectories,\u003C\u002Fli>\n\u003Cli>and expanding skill coverage beyond the original training distribution.\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Ch3>Plug-and-Play Transfer Across Agents\u003C\u002Fh3>\n\u003Cp>The resulting skill library can be directly injected into different base agents, enabling \u003Cstrong>strong-to-weak transfer\u003C\u002Fstrong> without retraining the underlying model.\u003C\u002Fp>\n\u003Ch3>Better Performance and Efficiency\u003C\u002Fh3>\n\u003Cp>SkillX consistently improves:\u003C\u002Fp>\n\u003Cul>\n\u003Cli>\u003Cstrong>task success rate\u003C\u002Fstrong> on challenging benchmarks,\u003C\u002Fli>\n\u003Cli>\u003Cstrong>execution efficiency\u003C\u002Fstrong> by reducing unne\u003C\u002Fli>\n\u003C\u002Ful>\n",1784240408124]