Safety message dissemination in a vehicular ad-hoc network (VANET) is in desperate need of vehicle-to-vehicle (V2V) communication with low latency and high reliability. The dynamics of vehicle passing and queueing as well as high mobility create distinctive propagation characteristics of wireless medium and inevitable uncertainty in space-time patterns of the vehicle density on a road. It is therefore of great importance to integrate stochastic geometry of vehicle locations into V2V channel modeling. In this paper, we characterize intervehicle communication in a random field of vehicles, where a beacon or head vehicle (transmitter) geobroadcasts safety or warning messages to neighboring client vehicles (receivers) randomly located in a cluster on the road. To account for a doubly stochastic spatial property of the VANET, we first model vehicle's random locations as a stationary Cox process with Fox's H-distributed random intensity (vehicle concentration) and derive the distributional functions of the 'th nearest client's distance from the beacon in such a Fox Cox field of vehicles. We then consolidate this spatial randomness of receiving vehicles into a path loss model and develop a triply-composite Fox channel model that combines key wireless propagation effects such as the distance-dependent path loss, large-scale fading (shadowing), and small-scale fading (multipath fading).