We present BlobSnake, a casual game designed to help generate new feature representations in the context of Human Activity Recognition. Feature selection is an essential task to be completed in the context of developing any non-trivial activity recognition system for a new set of activities. Presently, using anything other than a set of standard features requires a considerable amount of effort to be expended upon expert driven algorithm development. BlobSnake is an alternative approach which uses direct interaction with real sensor data by non-experts in order to develop additional features, thus lowering the cost and expertise otherwise required to produce more effective recognition performance. Our experiments demonstrate that our method improves upon the state of the art performance of standard features in a challenging recognition scenario.