I suspect the photocell ckt. had a cap open up. Check it in daylight. Block it with something opaque, not your hand, and time how long it takes for the light to come on. Remove the cover and observe how long it takes for the light to turn back off. They should both be substantial time delays. But you say you swapped with an identical fixture (known good???) and got the same results? You should still do the timing test on a non-flickering fixture as a basis for comparison.
Secondly, the CdS might be reacting to a non-visible light source. I don't know the spectral response off hand, just that the resistance varies from hundreds of ohms in strong light, to millions of ohms in darkness. A filter film over the window might fix that.
Photocell ckts are designed with hysteresis, also called positive feedback, snap action, differential, deadband, schmitt trigger action, and some I'm forgetting. If some of the light from the fixture that you're controlling, can reflect back to the photocell, you might be exceeding the switching threshold. Or could this light be interacting with another fixture controlled by a photocell??? I had to troubleshoot a complaint about the lights going off at a bus vaulting island a few years ago. I climbed the ladder and eyeballed the viewing window for the photocell. I could see the pretty new street light that the city installed across the street. The fix involved an empty pop can, tin snips and a couple cable ties. I blocked the sight line between the photocell and street light, while still allowing ambient light to enter the window.