Use only one row. Skeleton screens provide users a mental model that anticipates what users will see on a more or less similar way to what they will actually see, just not necessarily the same.
If you see the image below, you'll see that the skeleton placeholders aren't exactly the same. However, they're similar enough to convey what the final result will be:

Another example from this paper on skeleton screens

Remember that, while it's possible to replicate the structure exactly as it will be, it's not necessary, as long as you provide the general structure overview, it will be enough for the user to understand and anticipate what the screen will look like.
EDIT: I'll add this as an anecdote; please don't take it as research that applies to all cases.
Two years ago, we helped build the new interface for a big newspaper. We used skeleton screens, but then research showed something: speed was so fast that the skeleton screens showed as FOUC (Flash of Unexpected Content). However, since we obviously couldn't know whether the connection would be fast or not, we needed the skeleton screens. So, the solution was to artificially delay the load to avoid FOUC. Therefore, we had to worsen the experience to avoid an unexpected event that would result in (perhaps) an even worse experience.
Again, do not take this as something that applies to all cases, just providing some food for thought.