Two striatal subregions, the Dorsolateral Striatum (DLS) and the Dorsomedial Striatum (DMS) are thought to implement habitual and flexible, goal-directed behavior respectively. The DMS has been shown to be involved in tasks such as gating of information into working memory and accumulating evidence before making a decision. How the DMS produces these behaviors is less understood than the DLS because, unlike habitual motor tasks used to study the DLS, tasks used to study the DMS require more management of internal mental states, an inherently unobservable process. In this work, we build a computational model that can operationalize the abstract process by which the DMS produces flexible, goal-directed behavior, borrowing ideas about motor action selection in the DLS. Our central hypothesis is that, similarly to how the DLS interacts with a low-level cortical region, such as motor cortex, to support motor actions, the DMS interacts with a high-level cortical region, such as Prefrontal Cortex (PFC), to support mental actions that shape the activity of PFC to provide optimal task context for the DLS and motor cortex to act on. To instantiate this hypothesis, we expand on previous work (Mizes et al. 2023 NatNeuro) that use Recurrent Neural Networks (RNNs) to model how the DLS implements habitual motor behavior by reproducing the results of their network and augmenting the network with an additional corticostriatal loop (PFC+DMS) to enable flexible, goal-directed behavior. We train the PFC on a generic working memory task, fix its weights, and train the DMS to shape pre-trained PFC representations to quickly learn new tasks involving working memory gating and evidence accumulation, two goal-directed behaviors that the DMS has been implicated in. We discovered that DMS enables fast learning of these behaviors by reprogramming pretrained PFC representations to linearly separate either past gated stimuli or amount of accumulated evidence.