Recently, researchers proposed a framework called SkillShapley to estimate the contribution of skill steps in LLM agents. The framework operates in two phases: first, it identifies informative coalitional regions, and then adaptively samples new coalitions to obtain reusable marginal evidence. The experimental results show that SkillShapley can effectively identify high-value or low-value skill steps, which is significant for agent skill creation. The proposal of SkillShapley solves the limitation of existing agent skill creation methods, which lack quantification of individual step contributions. Through SkillShapley, researchers can better understand the impact of each step on overall skill performance, thereby optimizing the skill creation process. Blogger's Review: The SkillShapley framework provides a new perspective on skill step attribution, and through boundary-adaptive Shapley valuation, it can more accurately evaluate the contribution of each step, which is of great reference value for agent skill creation and optimization.