Can quantum entanglement be used to coordinate actions at "FTL speeds" without breaking causality or actual faster-than-light communication?

Going off of WillO's answer, while this scheme would work it would be no more effective than using a printer and two pieces of paper. Yes, your scheme is different in that it involves quantum nonlocality, but nevertheless it does not constitute faster-than-light communication because no information is being transferred between the two leaders. Their respective observations are correlated, but are nevertheless random. Hence, there's no problem. Is it weird? Yes. Is it a threat to causality? No. :)

Alternatively, they could have a computer print out two copies of the same random number, stuff the copies in their pockets, consult them when it's time to attack, and attack from one direction or another depending on whether the number is odd or even.

Why does this not constitute FTL communication? If the answer to that is clear to you, then you've answered your own question.

STL

The information is inherently contained in the particle pair themselves. Therefore, information only moves as quickly as the particles themselves do. What is not included in the description of your scenario is how the entangled particles originate. The particles cannot become entangled unless they are in very close proximity to begin with. Thereafter, they must move at slower-than-light speeds to their respective destinations. Even though the final state of the observable is not known until the observation occurs, I think it is fair to say that the information about the final state is implicitly encoded in the entanglement, rather than the observation.

In this sense, I think the paper printing scenario really is apt. Here, we would say that the same printer must print both pieces of paper, and the paper has the special property that once you look at it, the information is lost. This is exactly equivalent to saying that the information is not there until you look at it, from the perspective of an observer. That is, an observer cannot tell the difference between the printer deciding the outcome at the time of printing vs. the papers magically manifesting the same value at the time of observation. Even so, the printer analogy makes it clear that the information is created at "entanglement time" (i.e., print time), and thus, the information speed is simply the speed at which the papers move about, not the speed at which they are observed.