The core idea is to consider software-architecture evolution tasks as a knowledge that must be clearly modeled and properly managed. The main expected benefit is the reuse of existing and already available evolution expertise rather than reinventing it, sometimes awkwardly and thus avoid time-consuming redundant evolution activities. For this purpose, we propose to use the evolution style concept as a neutral interchange format to capitalize and transfer knowledge about domain-specific evolution tasks. In this paper we put the focus on how it is possible to reason on evolution-styles description libraries through a classification scheme. Specifically, we present the "evolution shelf'', an infrastructure to perform (a) incremental acquisition of new evolution descriptions and (b) retrieval of evolution descriptions matching with a given context. Our shelf, dedicated to software architects, relies on well-known repository techniques while updating them to support and exploit the evolution-style concept.