The advise I have seen generally amounts to what you are doing, mathematics is (even though I'll get stoned for saying this) an eminently experimental science. You play around with concepts, try to see how they fit together, stumble on a problem and look for ways to solve it. You find something that looks interesting, you try to prove it is true (by fooling around with it, seeing why it has to be true); you fail repeatedly and begin believing it really isn't (always) true, and start looking for ways in which it can fail.
The only way to become proficient in anything is to dedicate some 10.000 hours total to it. And much of those hours will be "wasted", in that you won't be able to show groundbreaking theorems, or even just correct results, for most of those hours. Just keep at it.
Read (and make sure you understand) Pólya's "How to solve it". There are other books by prominent mathematicians in which they discuss the soft(er) side of the profession, mostly in form of a scientific autobiography of sorts. I'm sure others around here can add a list of suggestions.
It is tacitly assumed that whoever does math is automatically a stellar writer. This just isn't true. Sure, there are people who are naturally gifted, but we the vast majority are mediocre or much worse at it. But it is a skill that can be improved. Look up e.g. Halmos' "How to write mathematics", Knuth et al's "Mathematical writing", there probably are many others worthy of a read.
Much mathematics is financed by teaching... and again, that isn't something many of us are born knowing how to do well (or at all). Try to get some training in that area. I've found out the hard way that doing it right involves lots of rather counterintuitive stuff. Many stumbling blocks for rank beginners are in areas we have so well travelled that we just can't see the snares and pitfalls anymore. Concepts we find simple on closer analysis aren't, or clash so blatantly with the common-sense meaning of the technical terms used to describe them that understanding is next to impossible.