I've stood over friends' shoulders and done similar live-coaching. It definitely helps.
You're making this sound easy, but in reality it's very complicated.
How to you measure how much someone is fighting? What if they're playing a very aggressive style? What if the opponent is turtling?
What is "too much" resource float? What time period should it be measured over? What if the player correctly prioritized more important tasks over spending resources?
What is a "big army" and what does "doing nothing" mean? What if they're playing a passive style? What if they're actively roaming the map but not fighting?
Etc.
I have tried to create an analysis tool that does similar things for StarCraft 2, and it's extremely difficult because of how variable and contextual everything is.
It's easy for humans to look at these things and interpret the situation. It's way harder to create rules that accurately reflect human interpretation.
I don't think it's hopeless though, the AI already does something similar. As a very naive implementation, you could for instance run your AI logic on the current state of the game and compare your move to the expected AI move. A bit like with a chess engine.
For instance in Age of Empires 2, the best way to improve is to look at a replay of your games, and analyse what went wrong. Stuff like "my economy was idle because I was focusing on the fighting" is much more flagrant then, during the game you often think it was just a few seconds, when in reality it could be a whole minute. I could see a lot of value in having an assistant that analyse your game after the fact and give you hints like: you focus too much on fighting, you have too much resource float, you have a big army but it's doing nothing, your army composition was subpar, you should have anticipated the switch to X, ... You could have it run real time as well, but then you get into the "is it cheating?" debate.
Action games are certainly more skill based as you said, and it's harder to give actionable advice.