README.md

👉 SkillX 👈

SkillX: Automatically Constructing Skill Knowledge Bases for Agents

👀Overview
🔧Installation
🏃Quick Start
🎁Acknowledgement
🚩Citation

📖 Overview

SkillX is a fully automated framework that constructs a reusable, plug-and-play skill knowledge base for LLM agents from experience.

Instead of storing raw trajectories, workflows, or loosely structured reflections, SkillX distills agent experience into a three-level skill hierarchy:

Planning Skills for high-level task organization
Functional Skills for reusable tool-based subroutines
Atomic Skills for execution-oriented tool usage patterns

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 AppWorld, BFCL-v3, and τ2-Bench, SkillX consistently improves both task success and execution efficiency.

Data Formats

Trajectory Input (JSONL)

SkillX expects trajectories in the following schema:

{
  "trajectory_id": "traj_001",
  "task_id": "task_001",
  "user_task": "How many songs are in my Spotify library?",
  "task_history": [
    {"role": "system", "content": "You are a helpful assistant..."},
    {"role": "assistant", "content": "I'll help you count..."},
    {"role": "user", "content": "Output:\n```\n{\"songs\": 150}\n```"}
  ],
  "reward": 1.0,
  "metadata": {}
}

🤖 Key Features

Hierarchical Multi-Level Skill Design

SkillX transforms raw trajectories into a structured three-tier skill space:

Planning Skills capture high-level decomposition and ordering
Functional Skills represent reusable multi-step tool subroutines
Atomic Skills encode practical tool usage constraints and patterns

Fully Automated Skill KB Construction

SkillX provides an end-to-end automated pipeline that:

rolls out agents on training tasks,
extracts reusable skills from successful trajectories,
consolidates and filters low-quality skills,
and builds a reusable plug-and-play skill knowledge base.

Iterative Skill Refinement

SkillX continuously improves the skill library through:

skill merging for consolidating redundant behaviors,
quality filtering for removing brittle or hallucinated skills,
and iterative updates that add, modify, or keep skills based on execution feedback.

Exploratory Skill Expansion

Beyond seed demonstrations, SkillX proactively discovers new skills by:

identifying under-used and failure-prone tools,
guiding environment exploration,
synthesizing new tasks from exploratory trajectories,
and expanding skill coverage beyond the original training distribution.

Plug-and-Play Transfer Across Agents

The resulting skill library can be directly injected into different base agents, enabling strong-to-weak transfer without retraining the underlying model.

Better Performance and Efficiency

SkillX consistently improves:

task success rate on challenging benchmarks,
execution efficiency by reducing unne

SkillX