A system where agents autonomously design, refine, and store task-specific skills as 'stateful prompts' to achieve non-parametric continual learning.
March 20, 2026
Original Paper
Memento-Skills: Let Agents Design Agents
arXiv · 2603.18743
The Takeaway
This moves away from the 'frozen weights' paradigm by externalizing intelligence into an evolving library of structured skills. It enables an agent to become more capable over time without ever updating the underlying LLM parameters.
From the abstract
We introduce \emph{Memento-Skills}, a generalist, continually-learnable LLM agent system that functions as an \emph{agent-designing agent}: it autonomously constructs, adapts, and improves task-specific agents through experience. The system is built on a memory-based reinforcement learning framework with \emph{stateful prompts}, where reusable skills (stored as structured markdown files) serve as persistent, evolving memory. These skills encode both behaviour and context, enabling the agent to c