Prompt execution of planned motor action is context-dependent and a critical component of animal behavior. Studies from humans, non-human primates, and rodents have ubiquitously established that the primary motor cortex (M1) is the main hub for skilled motor learning, which relies on the precise coordination of multiple brain regions. Yet, how various populations of neurons separated by millimeters reliably communicate is still unclear. Here, we investigate how moment-to-moment and trial-to-trial fluctuations of population dynamics in the secondary (M2) motor cortex relate to stable population codes in the primary motor cortex (M1) during a context-dependent volitional motor task. Using state-space analysis, we find that M2-M1 spiking trajectories occur through a communication subspace, suggesting that a small fraction of selective neural dimensions interact across motor areas, which changes as a function of task variables. Importantly, the stability of neural trajectories was accompanied by spatiotemporal LFPs organized as propagating traveling waves (TWs) recorded across the brain's surface. We posit that traveling waves enable patterned timing of inputs to specific brain regions thus facilitating communication within an optimal window of time. Indeed, the structure of traveling waves reflected task-relevant outcomes. Specifically, changes in wave phase directionality were correlated to the quenching of neural variability of cortical neurons, which were tightly coupled with ongoing wave dynamics within a lower dimensional state-space. Using focal cooling and optogenetic inhibition, we show that M2 modulates the structured generation of traveling waves and neural trajectories in M1 for correct motor execution via distinct pathways: cortical and trans thalamic. Thus, traveling waves may support the dynamic rerouting of neural dynamics within communication subspaces to guide optimal motor output.