Abstract: When a multi-agent system are executing global tasks, the behavior of an individual agent may lead to the entire system getting stuck in a deadlock state. To address this issue, this paper focuses on a heterogeneous multi-agent system where different agents can perform different tasks or accomplish the same task using different methods for dynamic global tasks. A novel supervisory-control framework is proposed to guarantee that the dynamic global task is completed safely and without deadlocks. First, each agent is modeled as an independent discrete event system, and model checking techniques are used to assess whether an agent has the capability to perform a specific task. Next, for each agent capable of executing a specific task, a supervision controller is designed to meet specific standards to ensure the task is completed without blocking. Finally, the theoretical results are applied to a multi-agent system to validate its effectiveness.