Generally, Multi-Agent Systems (MAS) are used to solve diagnostic problems; however, sometimes we have to diagnosis the MAS itself. As an actual fact, in many concrete cases, complex activities can be decomposed to simpler activities where each simple activity is executed by a particular agent that can be tangible (robot) or intangible (software). In short, the original complex activity is performed with the aid of a team including a set of interacting agents. The present paper proposes a decentralized monitoring methodology that conquers the centralized approach limitations (specially, the scalability property). Precisely, the Global Monitoring Plan (GMP) is split to set of smaller plans which are allocated to monitor agents. Usually, the GMP decomposition generates a set of sub-plans which are dependent. Consequently, all along the monitoring process, the interaction between monitor agents is required to get agent level outputs that are consistent at system level. The proposed approach allows the monitor agent to detect, dynamically, the set of agents involved in the interaction process. This set is called Dependency-Set. In addition, we propose a structured collaboration procedure between the agents belonging to the same Dependency-Set.