We construct a numerical model of emission from minijets, localized flows driven by magnetic reconnection inside Poynting-flux-dominated jets proposed to explain the ultrafast variability of blazars. The geometrical structure of the model consists of two wedge-like regions of relativistically flowing gas, separated by a stationary shock. The dynamics is based on solutions of relativistic magnetic reconnection with a guide field from Lyubarsky. Electron distributions in each region are chosen to match the pressure and density of the local plasma. Synchrotron emission from both regions is used to calculate Compton scattering, Compton drag and photon–photon opacity effects, with exact treatment of anisotropy and the Klein–Nishina regime. Radiative effects on plasma are taken into account, including the dependence of pressure on electron radiative losses and adiabatic heating of the flow decelerating under Compton drag. The results are applied to the 2006 July flare in the BL Lac object PKS 2155−304, with the aim of matching TeV flux measurements by the High Energy Stereoscopic System (HESS) with models that satisfy the variability constraints, while keeping X-ray emission below simultaneous Chandra observations. We find that models of isolated minijets with a significant guide field overproduce X-ray emission, and that we must take into account the radiative interaction of oppositely oriented minijets in order to achieve a high enough dominance by Comptonized TeV radiation. We argue that such interactions are likely to occur in a jet where there is substantial internal reconnection, producing a large number of misaligned minijets. Finally, we show that large jet magnetizations are indeed required to satisfy all observational constraints and that the effective Lorentz factor of the minijet plasma has to be larger than 50, in agreement with earlier one-zone estimates.