Focusing on on-demand collaborative scheduling for reconfigurable swarms with multimodal capabilities, this work proposes a hierarchical decision-making architecture driven by both environmental and task layers. The architecture enables orderly transitions between different swarm modalities in response to obstacle traversal and fault events. A plug-and-play low-level controller library is developed to support the robust generation of task-sliced modal behaviours in complex scenarios. Two representative obstacle-crossing scenarios are presented to demonstrate the advantages of reconfigurable swarms in terms of on-demand reconfiguration and timely dispersion when facing various unexpected situations.