WikiSkill: Compiling Agent Experience into Persistent Knowledge for Skill Evolution
WikiSkill’s core claim is that agent skills improve when past execution experience is distilled into a persistent wiki.
The paper separates raw runs, accumulated knowledge, and executable skills, then uses the wiki to guide later skill updates. In benchmarks across models, the authors report stronger results than prior skill-evolution methods and most no-skill baselines. They also say evolved skills can transfer across model families, and that skills evolved by other models can beat self-evolved ones. ArXiv · AI/CL/LG's note
The paper separates raw runs, accumulated knowledge, and executable skills, then uses the wiki to guide later skill updates. In benchmarks across models, the authors report stronger results than prior skill-evolution methods and most no-skill baselines. They also say evolved skills can transfer across model families, and that skills evolved by other models can beat self-evolved ones. ArXiv · AI/CL/LG's note
score 5