MUSE-Autoskill: Self-Evolving Agents via Skill Creation, Memory, Management, and Evaluation

Large language model (LLM) agents rely on reusable skills to solve complex tasks. However, existing skill creation approaches treat skills as isolated and static artifacts, limiting their reusability, reliability, and long-term improvement. We propose MUSE-Autoskill Agent (Memory-Utilizing Skill Evolution), a skill-centric agent framework that lets agents continuously improve their task-solving capability by creating, reusing, and refining skills under a unified lifecycle (creation, memory, management, evaluation, and refinement). Our framework enables agents to create skills on demand, store and reuse them across tasks, organize and select them efficiently, and evaluate them through unit tests and runtime feedback for continuous refinement. We further introduce skill-level memory that accumulates experience for each skill across tasks, enabling more effective reuse and adaptation over time. Experiments on SkillsBench provide initial evidence that lifecycle-managed skills can improve task success, efficiency, reuse, and cross-agent transfer, highlighting the importance of treating skills as long-lived, experience-aware, and testable assets.

📋 한 줄 요약

**[LLM Agent / Self-Evolving Skills]** MUSE-Autoskill가 skill을 creation·memory·management·evaluation·refinement의 unified lifecycle로 다룸 — skill-level memory로 경험 누적, 단위 테스트·런타임 피드백 기반 평가, SkillsBench에서 task 성공률·효율·재사용·cross-agent transfer 모두 개선.

🎯 핵심 기여도

LLM agent가 complex task 해결에 reusable skill에 의존하지만, 기존 skill creation 접근이 skill을 isolated·static artifact로 다뤄 reusability·reliability·long-term improvement를 limit 함을 진단.
MUSE-Autoskill Agent (Memory-Utilizing Skill Evolution) 제안 — skill-centric agent 프레임워크로 unified lifecycle(creation·memory·management·evaluation·refinement) 하에 skill 생성·재사용·refinement로 task-solving 능력 continuously 개선.
Agent가 on demand로 skill 생성, task 간 저장·재사용, 효율적 organize·select, unit test·runtime feedback으로 평가해 continuous refinement.
Skill-level memory 도입 — skill별 task 간 경험 누적, 시간에 걸친 more effective reuse·adaptation 가능.

💡 핵심 아이디어

LLM agent의 skill을 isolated artifact가 아닌 long-lived·experience-aware·testable asset으로 다뤄야 하며, creation→memory→management→evaluation→refinement의 unified lifecycle과 skill-level memory가 결합되면 task 성공률·효율·재사용·cross-agent transfer를 동시 개선한다.

🔬 기술적 접근법

**방법론**: MUSE-Autoskill — 5-stage skill lifecycle + skill-level memory.
**핵심 기법**: (1) On-demand skill 생성, (2) Task 간 저장·재사용 메커니즘, (3) 효율적 skill organization·selection, (4) Unit test + runtime feedback 기반 평가로 continuous refinement, (5) Skill-level memory가 각 skill의 task 간 경험을 누적해 시간에 걸친 reuse·adaptation 강화.

📊 주요 결과

SkillsBench 실험으로 lifecycle-managed skill이 task 성공률·효율·재사용·cross-agent transfer 개선.
Skill을 long-lived·experience-aware·testable asset으로 다루는 것의 중요성 강조.

💭 의의 및 한계

**의의**: Skill을 isolated artifact가 아닌 lifecycle managed asset으로 재정의, skill-level memory라는 새 추상화, unit test + runtime feedback의 evaluation 결합, cross-agent transfer 가능성 입증. **한계**: SkillsBench 단일 평가로 다른 도메인 일반화 검증 추가 필요, lifecycle 5-stage의 운영 overhead, skill 평가의 unit test 작성 부담.

🚀 실용적 활용

자율 LLM agent의 skill 자가 개선.
Multi-agent 시스템 간 skill 공유·transfer.
Continuous learning agent 인프라.