Memento-Skills is an agent framework that allows deployed agents to autonomously improve by learning, rewriting, and evolving their own skills. The system uses a reflection loop for self-evolution and supports multiple open-source LLM ecosystems including Kimi, MiniMax, and GLM. It provides both CLI and GUI interfaces with a local sandbox execution environment and integrates with various IM platforms like Feishu, DingTalk, WeCom, and WeChat.