◇ Could this help me?
Yes. The one idea worth extracting is in step two: handing an agent a reference design along with what the thing is actually for produces adaptation rather than imitation, and that distinction is most of the difference between a dashboard that looks generated and one that looks designed. Adding a custom banner image is a small, real trick for the same reason, since the generic parts of AI-produced interfaces are what make them recognizable. Two things to handle more carefully than the video does: a skill is code you are installing and deserves a look at its publisher first, and publishing live should follow a check of what the dashboard actually exposes. General practice.
I am building a dashboard and want it to look designed rather than generated. Here is a reference design I like and here is what my dashboard is actually for, including who uses it and what decision they make with it. Adapt the reference to my purpose rather than reproducing it: tell me which parts of the reference do not serve my use case and should be dropped, and what my case needs that the reference does not have. Then build it. Before we publish anything, list every piece of data the dashboard exposes and tell me which of it should not be publicly reachable.