The rise of livestreaming as a new sales mode has greatly facilitated the thriving development of China's e-commerce platforms, where the market scale of livestreaming reached 4.9 trillion RMB in 2023. However, it may lead to significant uncertainty in product variety and volume in a short term, e.g., “Double Eleven” shopping festival, posing challenges to the responsiveness of manufacturers. Seru production systems (SPSs) can cope with volatile markets and have gained benefits for many Asian electronics manufacturers. As a human-centered system in high-cost high-skill labor environments, workers in an SPS are commonly partially cross-trained and exhibit diversity in their skill sets and skill levels. This study explores the influence of demand uncertainty and worker skill diversity on SPS management, a topic that is rarely investigated in the literature despite its importance. Theoretical conditions on worker skills are examined to enable a partially cross-trained seru to achieve performance equivalent to that of a fully cross-trained seru, which reveals how to utilize the skill flexibility within a single seru. Then, from a macro to micro perspective, a joint problem of seru formation, order assignment, and workload assignment is formulated as a two-stage stochastic mixed-integer programming model. An exact algorithm based on Benders decomposition is designed and enhanced by a heuristic local search inspired by analytical results. Numerical results confirm the necessity of considering demand uncertainty to operate a flexible and profitable SPS and provide suggestions for cross-training in various production environments.