A fundamental problem in neuroscience is how neuronal activity in the brain generates organized and stable behaviors across multiple timescales. Recent research in many model organisms, e.g. in monkeys, led to the observations that the neuronal encoding of individual actions, such as preparation for and execution of arm movements, is achieved by large neuronal populations but often lies on low-dimensional manifolds. However, it is largely unclear if such manifolds also coordinate higher dimensional activity patterns that correspond to simultaneously executed fine-tuned movements. In order to address these questions, we established a whole brain single-cell resolution recording and analysis pipeline in freely crawling C. elegans nematodes, which display natural long time-scale behavior sequences including switches between different actions. We observed a structured manifold resulting from brain-wide coordinated neuronal activity patterns. Here, we developed a principled method for investigating the relationship between the global manifold and any residual higher dimensional activity patterns as well as the behavioral correlates thereof. We first show that the global manifold encodes a high-level action sequence. We show that most of these neuronal classes are involved in generating intrinsic and distributed motor commands. However, a fraction of neuron classes rather encode re-afferent sensory perception that results from the execution of actions. Specifically, we observe that a single sensory modality (O2 and CO2 sensation) accounts for ~12% of these neurons. We additionally find that the manifold hierarchically organizes individual neural activity corresponding to behavioral encoding. Mathematically, we use a multilevel Bayesian model to find neurons whose activity reflects behavioral encoding that is not uniform across time, but is gated by the manifold. Thus, the manifold is a composition of both globally shared internal action commands which organizes local activity, and re-afferent sensory information. The mix of sensory and command signals highlight the need for detailed knowledge and additional experiments to properly interpret potential causal links between neuronal activity and behavior. In conclusion, we propose that neuronal manifolds provide brain-wide shared context about the current behavioral state to orchestrate local movement patterns and to process movement-related sensory information.