Continually Evolving Skill Knowledge in Vision Language Action Model