Memento-Skills: Let Agents Design Agents

Dimensions overall score 8.0

• Memento-Skills achieves a 26.2% relative improvement in overall accuracy on the General AI Assistants benchmark.result 95%
• Memento-Skills achieves a 116.2% relative improvement in overall accuracy on Humanity's Last Exam.result 95%
• Memento-Skills enables continual learning without updating LLM parameters through evolution of externalized skills and prompts.technical 90%
• The system uses a Read-Write Reflective Learning mechanism where a skill router selects relevant skills in the read phase and updates the skill library in the write phase.method 90%
• Memento-Skills enables a generalist agent to design agents end-to-end for new tasks, unlike prior approaches that rely on human-designed agents.technical 85%
• The approach may rely heavily on the initial quality of skill definitions and the ability to generalize beyond predefined skill sets.limitation 80%
• Memento-Skills uses skills stored as structured markdown files that serve as persistent, evolving memory.method 90%
• Scalability and interpretability of autonomous decisions might pose challenges for Memento-Skills.limitation 75%

Competitive landscape

Competitor map is still being generated for this paper. Enable generation or check back soon.

Startup potential card