Flexible integration of sensory information in a context-dependent manner is a key cognitive process required to generate appropriate behavior. An intriguing question, then, is how the same sensory stimulus can be interpreted differently according to context in order to generate different behavioral responses. We designed a task in which mice were trained to lick for reward in response to a brief single whisker stimulus if it was preceded by a brief Go-Tone presented one second before the whisker stimulus, but not if it was preceded by a NoGo-Tone. Optogenetic inactivation of primary whisker somatosensory cortex (wS1), secondary whisker somatosensory cortex (wS2), secondary whisker motor cortex (wM2), or anterior lateral motor cortex (ALM) during the presentation of the whisker stimulus strongly decreased the probability of licking in the reward window in Go-trials. Inactivation of wM2 and ALM during the delay between the Go-tone and the whisker stimulus also reduced licking in the reward window. We recorded neuronal activity using multiple Neuropixels probes simultaneously. Prominent persistent activity following the Go-tone presentation was found selectively in wM2 and ALM, even in trials devoid of delay period movements. Using linear decoding of neuronal activity, we found that the accuracy of classifying context in the 200 ms before the whisker stimulus is significantly higher than the baseline chance level, with the highest accuracy in wM2. These findings suggest a crucial role of the frontal areas wM2 and ALM in the encoding and maintenance of contextual information in a short-term memory task.