The coordination of neural activity across brain areas during a specific behavior is often interpreted as neural communication involved in controlling the behavior. However, whether information relevant to the behavior is actually transferred between areas is often untested. Here, we developed and used information theoretic tools to identify, in neural activity, encoding and transmission of information relevant to behavior. We then used these new methods to quantify how motor cortex and striatum encode and exchange behaviorally relevant information about specific reach-to-grasp movement features during skill learning. We found a temporal shift in the encoding of behaviorally relevant information during skill learning, as well as a shift in the direction of behaviorally relevant information flow, from cortex-to-striatum during naive movements to striatum-to-cortex during skilled movements. Standard analytical methods that quantify the evolution of overall neural activity during learning – such as changes in neural signal amplitude or the overall exchange of information between areas – could not capture these behaviorally relevant information dynamics. Using standard methods that, in contrast to our developed method, are insensitive to the behavioral-relevance of transmitted information, we found a coactivation of overall neural signals during movement production and an increase in overall information propagation between areas during learning without a detectable shift in directionality of communication. We conclude that skill learning relies on re-routing of behaviorally relevant information across cortex and striatum, and that isolating the neural activity informative about behavior is critical to uncover directional interactions within a coactive and coordinated network.