Dynamically adding sensors to the Extended Classifier System (XCS) during its learning process in multistep problems has been demonstrated feasible by using messy coding (XCSm) and s-expressions (XCSL) as the representation of classifier conditions. XCSm and XCSL shown improved performance when new sensors were dynamically added to the agent of these systems in addition to the original available sensors during the learning process. However, these systems may suffer from overspecified problem and some logical operators (or clauses) could lead instability of the performance. Despite studies have suggested that these issues can be solved by appropriate parameter tuning, in the last study, we introduced a novel representation of classifier conditions for the XCS, named Sensory Tag (ST) (called XCS with ST condition, XCSSTC) to achieve the same goal as XCSm and XCSL, but inherent most of the mechanisms of the XCS to solve those issues that the XCSm and XCSL encountered without any parameter tuning. The experiments of the proposed method were conducted in the multistep problems (i.e. Woods1 and Maze4). The results indicate that the XCSSTC is capable of being dynamic added additional sensors to improve performance during the learning process, and moreover, the XCSSTC shown a better performance in regard to learning speed than the other methods.