MetaSkill-Evolve: Recursive Self-Improvement of LLM Agents via Two-Timescale Meta-Skill Evolution
2026-07-06 • Artificial Intelligence
Artificial Intelligence
AI summaryⓘ
The authors study ways to help AI agents improve their skills on tasks by rewriting how they do the tasks themselves. They build a system called MetaSkill-Evolve that not only improves the task skills quickly but also gradually improves the method used to improve those skills. This means the AI can enhance both what it does and how it learns to do better over time. Their approach, tested on three different benchmarks, shows better accuracy than methods that do not improve or only improve task skills. They achieve this without needing extra models or objectives beyond their main setup.
LLM agentsskill evolutionrecursive improvementtask skillsmeta-skillsAnalyzerRetrieverAllocatorProposerEvolver
Authors
Zefeng Wang, Minxi Yan, Jinhe Bi, Sikuan Yan, Volker Tresp, Yunpu Ma
Abstract
Recent LLM agents tackle increasingly long-horizon, open-ended tasks, and external skills, reusable procedural knowledge supplied to the agent, further extend this capability. However, a fixed, hand-authored skill is rarely optimal, and cannot adapt to the diversity of tasks an agent encounters. Self-improving agents address this by rewriting their own skill files from execution traces, yielding meaningful gains on challenging benchmarks. Yet such self-evolution remains non-recursive: it improves only the task skill (what the agent does) while the improvement procedure (how it improves) is authored once and held fixed. We introduce MetaSkill-Evolve, a two-timescale framework that makes agentic skill improvement recursive: every branch carries both a task skill $s$ and a branch-local meta-skill $m=(ψ,σ,α,π,\varepsilon)$ whose five components parameterise the Analyzer, Retriever, Allocator, Proposer, and Evolver agents of the improvement pipeline. Task skills evolve on a fast loop while the meta-skill evolves on a slower one under the same pipeline applied to itself, with no additional model or objective. With all five pipeline agents sharing a single frozen backbone, MetaSkill-Evolve outperforms no-skill, static-skill, and single-level evolution baselines on three agentic benchmarks (OfficeQA, SealQA, ALFWorld), improving held-out test accuracy over the raw backbone by +23.54, +16.09, and +1.92 points respectively.