一句话：实证刻画 GitHub Actions 在 CI/CD 中从确定性执行向协调与决策逻辑分层的演化模式。
CI/CD 流水线传统上被设计为确定性工作流，依据仓库事件执行预定义步骤，常通过 GitHub Actions 实现。随着项目演化，开发者会添加、更新或替换 Action，以应对需求变化、失败和环境差异，但学界对流水线演化过程中决策逻辑位于何处、以及确定性执行与协调/判断之间如何平衡仍缺乏了解。本文在 Action 层面对 CI/CD 工作流进行纵向分析，识别反复出现的转换模式，并采用分层视角——确定性执行、跨系统协调和决策导向逻辑——考察项目是收敛到稳定配置，还是日益嵌入决策与协调逻辑。研究提供关于流水线自动化局限的实证洞见，并指导将执行与高层决策更好分离的设计。
Continuous integration and deployment (CI/CD) pipelines are traditionally deterministic workflows that execute predefined steps in response to repository events, often implemented via GitHub Actions. As projects evolve, developers frequently modify workflows (adding, updating, or replacing Actions) to handle changing requirements, failures, and environmental variability. However, little is known about how pipelines evolve regarding where decision-making resides and how they balance deterministic execution with coordination and judgment. This study characterizes the longitudinal evolution of CI/CD workflows at the Action level, identifies recurring transition patterns, and examines whether projects converge toward stable configurations or increasingly embed decision and coordination logic. We analyze pipelines through a layered lens: deterministic execution, cross-system coordination, and decision-oriented logic, to provide empirical insights into the limits of pipeline automation and guide designs that better separate execution from higher-level decision-making.