Memento-Skills framework enables AI agents to learn and adapt autonomously without expensive LLM retraining

Memento-Skills, developed by multi-university researchers, lets AI agents continuously improve their capabilities through external memory rather than model retraining

The framework allows agents to update and expand their skills based on environmental feedback, reducing operational costs compared to fine-tuning model weights

Co-author Jun Wang positions the solution as an advancement over existing tools like OpenClaw and Claude Code for enterprise deployment of autonomous agents