MemSkill: Learning and Evolving Memory Skills for Self-Evolving Agents

Haozhen Zhang, Quanyu Long, Jianzhu Bao, Tao Feng, Weizhi Zhang, Haodong Yue, Wenya Wang · 2026

Read MemSkill: Learning and Evolving Memory Skills for Self-Evolving Agents by Haozhen Zhang, Quanyu Long, Jianzhu Bao, Tao Feng, Weizhi Zhang, Haodong Yue, Wenya Wang (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-02474.txt.

Description

Haozhen Zhang and 6 co-authors let an AI agent's memory procedures themselves learn and evolve, in a 2026 arXiv paper. Most agent memory systems rely on a few static, hand-designed operations for storing and revising information. Those fixed routines hard-code human guesses and cope poorly with diverse interactions or long histories.

MemSkill recasts the operations as reusable memory skills for extracting, consolidating, and pruning information. A controller learns to select a small set of relevant skills, and an executor model carries them out. A third component, the designer, reviews hard cases where memories came out wrong or incomplete and proposes refinements or entirely new skills. The loop improves both selection and the skill set itself. Experiments on conversation, question-answering, and household-task benchmarks beat strong baselines and generalized across settings.

Catalog record

CREATORHaozhen Zhang, Quanyu Long, Jianzhu Bao, Tao Feng, Weizhi Zhang, Haodong Yue, Wenya Wang
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE70 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602-02474

Browse the full library · About the Tower of Babel