AutoSkill: Experience-Driven Lifelong Learning via Skill Self-Evolution

English | 中文

AutoSkill is a practical implementation of Experience-driven Lifelong Learning (ELL). It learns from real interaction experience (dialogue + agents), automatically creates reusable Skills, and continuously evolves existing Skills through merge + version updates.

News

• 2026-05-09: Added the installable AutoSkill Local Skill Manager (skills/autoskill) for maintaining local Agent Skill files after sessions, including reusable-experience triage, similar-skill search, and discard / improve / merge / create decisions.
• 2026-03-23: SkillEvo 1.0 released (Enabling models to iteratively self-evolve Skills through replay, evaluation, mutation, and promotion).
• 2026-03-13: AutoSkill4Doc 1.0 released (Being expext by extracting skills from document/research paper).
• 2026-03-01: Added offline skill extraction from archived conversations (See Skills in SkillBank/CovSkill).
• 2025-02-26: AutoSkill4OpenClaw 1.0 released (Extracting skills from trajectory of OpenClaw).
• 2025-02-04: AutoSkill 1.0 released (Extracting skills from dialogues in time).

Table of Contents

• News
• 1. Project Overview
• 2. Main Components
• 3. Skill Lifecycle Example
• 4. Documentation Map
• 5. Repository Structure (Top Level)
• 6. Star History
• 7. Citation
• 8. Contributions and Acknowledgments

1. Project Overview

• Experience-driven continuous skill evolution: extracts reusable capabilities directly from real user interactions and agent traces, then continuously maintains versioned skills so the system aligns with user needs over time.
• Universal skill format: uses the Agent Skill artifact (SKILL.md) with explainability and editability. Skills remain readable, reviewable, and manually revisable.
• Offline extraction from completed data: existing chats and trajectories can be imported directly for offline skill extraction; there is no need to replay the original interaction.
• Long-term capability value: AutoSkill turns short-term interaction signals into long-term capability assets that can be reused across runtimes.

2. Main Components

• autoskill/: core SDK, Web UI, OpenAI-compatible proxy, online skill evolution, offline conversation extraction, and offline trajectory extraction.
• AutoSkill4Doc/: standalone document-to-skill pipeline for extracting reusable skills from papers, manuals, and domain documents.
• AutoSkill4OpenClaw/: OpenClaw integration for trajectory-driven skill evolution and native skill mirroring.
• SkillEvo/: replay, evaluation, mutation, and promotion framework for iterative skill self-evolution.

3. Skill Lifecycle Example

A) Auto Decision + Feedback-triggered Extraction & Skill Management (v0.1.0)

If the user only asks to "write a report" and gives no stable preference/correction, AutoSkill does not create a new skill (it outputs an empty extraction result) to avoid noisy, generic skills.