New framework lets AI agents rewrite their own skills without retraining the underlying model

Memento-Skills is a framework that lets agents self-develop skills without retraining the underlying model.

It provides an evolving external memory scaffold to enable continual learning and progressive capability improvement.

Agents can update and expand their skills based on feedback from their environment.

For enterprise agents, avoiding retraining and manual skill building lowers operational overhead and data needs.

The framework can add continual-learning capability to existing offerings such as OpenClaw and Claude Code.