The automation of application management is one of the most important issues in Cloud Computing. However, the steadily increasing number of different services and software components employed in composite Cloud applications leads to a higher risk of unexpected side effects when different technologies work together that bring their own proprietary management APIs. Due to unknown dependencies and the increasing diversity and heterogeneity of employed technologies, even small management tasks on single components may compromise the whole application functionality for reasons that are neither expected nor obvious to non-experts. In this paper, we tackle these issues by introducing a method that enables detecting and correcting unintended effects of management tasks in advance by analyzing the context in which tasks are executed. We validate the method practically and show how context-aware expert management knowledge can be applied fully automatically to running Cloud applications.