Paper 2602.02474
MemSkill: Learning and Evolving Memory Skills for Self-Evolving Agents
- Published
- Feb 2026
- Research lab
- Independent
- Citations
- 92
- GitHub
- 556 stars
01 In brief
Summary
MemSkill reframes LLM agent memory operations as learnable, evolvable memory skills.
It maintains a shared skill bank where each skill provides structured guidance (purpose, when to use, how to apply, constraints) for extracting, consolidating, or revising memories.
A controller, trained with reinforcement learning (PPO), selects a Top-K set of relevant skills per text span, and an LLM-based executor applies them in one pass to produce skill-guided memory updates.
A designer periodically mines hard cases from a sliding buffer, clusters them, and uses an LLM to refine existing skills and propose new ones, forming a closed-loop that alternates between learning skill selection and evolving the skill bank.
Experiments on LoCoMo, LongMemEval, HotpotQA, and ALFWorld show MemSkill consistently outperforms strong baselines (e.g., MemoryOS, A-MEM, LightMem) in F1, LLM-judge scores, and success rates, while reducing LLM calls and interaction steps.
It also generalizes across base models (LLaMA to Qwen) and datasets (LoCoMo to LongMemEval, HotpotQA) without retraining.
Ablations confirm both the learned controller and the designer's skill evolution contribute to performance.
Evolved skills show domain specialization, e.g., temporal/activity details for dialogues and action constraints/object locations for embodied tasks.
02 From the paper
Abstract
Most Large Language Model (LLM) agent memory systems rely on a small set of static, hand-designed operations for extracting memory. These fixed procedures hard-code human priors about what to store and how to revise memory, making them rigid under diverse interaction patterns and inefficient on long histories. To this end, we present \textbf{MemSkill}, which reframes these operations as learnable and evolvable memory skills, structured and reusable routines for extracting, consolidating, and pruning information from interaction traces. Inspired by the design philosophy of agent skills, MemSkill employs a \emph{controller} that learns to select a small set of relevant skills, paired with an LLM-based \emph{executor} that produces skill-guided memories. Beyond learning skill selection, MemSkill introduces a \emph{designer} that periodically reviews hard cases where selected skills yield incorrect or incomplete memories, and evolves the skill set by proposing refinements and new skills. Together, MemSkill forms a closed-loop procedure that improves both the skill-selection policy and the skill set itself. Experiments on LoCoMo, LongMemEval, HotpotQA, and ALFWorld demonstrate that MemSkill improves task performance over strong baselines and generalizes well across settings. Further analyses shed light on how skills evolve, offering insights toward more adaptive, self-evolving memory management for LLM agents.