Detecting the current task of a user is essential for providing her with contextualized and personalized support, and using Contextual Attention Metadata (CAM) can help doing so. Some recent approaches propose to perform automatic user task detection by means of task classifiers using such metadata. In this paper, we show that good results can be achieved by training such classifiers offline on CAM gathered in laboratory settings. We also isolate a combination of metadata features that present a significantly better discriminative power than classical ones.