Reconfigurable intelligent surfaces (RISs) have been recently considered in sensing context to either improve localization performance or to extend the coverage in non-line-of-sight scenarios. While most of the literature considers RISs as statically placed in the environment, the usage of target-lodged RISs is relatively new and could be of interest where the target's reflectivity can/must be increased to improve its detection or parameters' estimation. This letter derives the Cram\'er-Rao bound (CRB) on the estimation of position and orientation (\textit{pose}) of a target-mounted RIS, in generic conditions: near-field, bistatic and wideband operation (i.e., when the wavefront across the RIS is curved and the employed sensing bandwidth is large enough to obtain a frequency-dependent RIS behavior). In particular, we focus on the wideband effect, that implies a \textit{pose-dependent filtering} on the impinging signal decreasing or increasing the CRB depending on the RIS size and the employed signal bandwidth.