> I'm afraid of accidentally running something on wrong gcp project / cluster combination out of mistake which will inevitably happen

I think GitOps as a philosophy helps here a bit. All changes to your cluster(s) go through git with a PR / review. Generally speaking you wouldn't be running kubectl or any CLI tools on your dev box where you hope you're working against the correct cluster or namespace.

I also like the idea of having a temporary test cluster that you routinely spin up and destroy for testing any type of Kubernetes upgrades or cluster-wide changes. This way you get a real result of how things will pan out before performing it on an important cluster.