This paper is concerned with representations which enable a technical agent to recognize objects and aggregates from mechanical parts as they evolve in an ongoing construction task. The general goal is that the technical agent has a detailed understanding of the task situation such that it can execute instructions issued by a human user in a dynamically changing situation. A levelled approach for comprehensive shape representation is presented which is motivated by a cognitive model of pictorial shape representations (prototypes). In particular, our system is able to derive and represent spatial properties (such as orientation) and geometric features (e.g., axes or planes) that can be ascribed to the developing construct.