Science has left the confides of philosophy a few centuries ago. However, science is an epistemology, and if you mean it in that sense, then 'philosophy' defines science.
The very scientific method is a philosophical theory. Philosophy works on the problems of how we know what we know, what constitutes good scientific theory, and what is valid reasoning.
Scientific reasoning, I think, is a special case of philosophical reasoning. Whereas in general philosophy one is not necessarily bound by the production of empirical observables, science (as an epistemology) makes empirical observations the key criteria for model success.
It may seem that the scientific method is an established technique, but it is a fluid one. The interpretation of the experiments of Bell's Inequality show us the great difficulty of applying the pedestrian versions of perception and knowledge. Philosophical theorizing is the only way out.
Science utilizes philosophical reasoning, but only as a means to find a basis to move forward into model creation and, hence, the production of observables such that the theory can be tested. Philosophers, I believe, can and should have an input into this process, however if nothing can come of it in the way of the production of observables, it is much ado about nothing from a scientific perspective. Without the production of observables such that specific models can be clearly validated or falsified in an empirical manner, the field remains within the confides of the philosophy of science with less relevancy to scientific research.
I think you are alluding to the far side of philosophy, that which brushes against theology. Philosophy's job is to distinguish between belief and knowledge. If that doesn't sound easy, it's because it is very difficult.
I was referring to philosophy as a tool of scientific reasoning that aids in theory selection. This can include pragmatic, aesthetic, logical, epistemological, etc, features of theory selection that provide a basis for selecting one model over another even if current observables are showing no particular model favorites. For example, the choice between QLG (quantum loop geometry) and string theories is largely a matter of philosophical prejudices. From a purely empirical stance, neither has much advantage over the other, so we have to utilize our philosophical prejudices in producing models that play to our prejudices with the hope of providing observables that empirical science will validate as being an adequate model of science.