The contribution deals with a class of intelligent decentralized systems that meet the agent paradigm. Such systems are marked by the possibility of arising critical situations, interpreted here as the threat of loss (partial or complete) of the system functionality. The work is focused on designing an overall architecture of the (sub-) system dedicated to the discovery of crises and support of anti-crisis activities. The architecture is proposed as a reference one so as it is possible to adjust it to the specificity of any particular application. As an illustration the case of a transportation system is discussed.