Systematically accumulating agent experience into a shared knowledge base makes skills more effective and transferable—smaller models with evolved skills can outperform much larger models without them.
WikiSkill is a framework that helps AI agents build and reuse skills by maintaining a persistent knowledge base (wiki) that captures insights from past experiences. Instead of letting optimization histories scatter across runs, the system consolidates what agents learn into organized, reusable skills that improve performance across different tasks and models.