The feedback loop occurs within the Sziklai pair itself. Assume, for simplicity, that the input voltage is constant. A change in the collector current of Q3, acting across R2, causes the emitter voltage of Q6 to change. Since the input voltage is constant, this device operates in common base mode, so the change in Q6 emitter voltage causes a change in its collector current. But Q6's collector current is Q3's base current, so the change in Q6's collector current is amplified by Q3 and appears at its collector, completing the feedback loop.
At DC the sense of this feedback is negative. Q3's fT provides the dominant pole at the frequency where its beta starts to fall from its DC value. If the loop gain has not dropped to less than 0dB before the additional phase shift from Q6 reaches 90˚, the loop will oscillate. Too fast a device in the Q3 position, relative to Q6, is likely to give trouble.