And yet you want to make a decision which consequences will largely depend on that.
It can be balanced at any point. The overall "cost" of a developed character is more or less the same - you just pay more for some skills and less for others.
I have played enough of such "balanced later" games. The result is usually terminally dull cavalcade of nerfs and arbitrary restrictions.
I don't think it can be done in any other way for me.
I have listed one.
Of course the more PnP mechanics replaced with gameplay mechanics - the better for CRPG, but it also means more money involved. Creating a whole separate mechanics for hacking and lockpicking is more taxing than just introducing a couple of skill checks.
The difference is much more fundamental. PnP mechanics is about a bunch of disjoint, lightweight (they need to be worked out fast by just a bunch of nerds with a pencil) arbitration mechanics stuck into flexible, human driven narrative (it is comparatively flexible and human driven even in a canned dungeon adventure, DMs don't throw syntax errors if you do something they didn't predict and you can attempt to do anything communicable in the language).
cRPG mechanics can be almost arbitrarily computation heavy, but must be well integrated and all-encompassing - what isn't accounted for by either basic mechanics or specific case-by-case scripting doesn't exist. For that reason they are also best built of possibly basic building blocks interacting with each other.
You have a massive variety of skills.
So far I'm looking in the opposite direction. More skills means more dump skills. That is of course depending on what do you call "massive". I'd rather have less skills that do more than more skills that do less. The prime example of this are speech skills, which are usually separated into persuasion, haggle and intimidation and two latter tend to be useless. Why not just combine it into into one skill (persuasion) and make others (haggle and intimidation) as perks for the main skill?
Then you should go the opposite direction regarding basics of your system as well. Coarse attributes controlling fine-grained skills is good when you can force player to commit to a rough build without knowing exactly how they will want to fine sculpt it in the end. If build variety is low because of few skills comparable to the number of attributes (or for n attributes per skill - number of attributes to the n-th power), your build system instead becomes an exercise in looking up optimal build(s) in the wiki - in other words shit.
OTOH making a good build system is effectively an exercise in leveraging combinatorics (possibly few attributes/skills/etc. AKA work in, possibly many differently playing builds AKA gameplay out), and making individual components of a build as orthogonal to each other as possible achieves just that and scales down to low number of skills just as well as up.
Merging or splitting skills/attributes depending on their relative utility is good practice but be aware of its impact and maybe try to minimize it.
Not sure if this will ever work or even need to. All skills can be equally useful, but you can't solve all problems by using only one.
If "fight, sneak, talk" are simple solutions I'd like to hear what is complex. So far modern "CRPGs" tend to forget even this.
Then your system is poorly suited to what you think you can do. Also, most modern and classic cRPGs are irredeemably shit. Even good ones are often floated by 1-2 elements, often created by accident and usually undermined by all the awful ones.
I want to ensure player that any skill will be useful, but also that he cannot solve everything by using only one. Some situations are unavoidable.
That's good, but if player can decide their final build just as they distribute their initial attribute points, then the extra gameplay provided by having to refine your build as the game progresses goes out of the window.
But it doesn't have to be realistic as long as it works in a game.
So what you say is that you wouldn't mind having attributes and things they control randomly (say, intelligence determining carry weight), because it's just a game?
At the very least attribute names are a way of communicating concepts to the player. It pays off to communicate clearly.
Games, especially modern ones tend to forget that and spew oceans of non-meaning through their stat screens and combat logs.
That's the main idea anyway. I was thinking about a connection between skills and attributes as a sort of a side effect.
This is a side effect that will require extra work to implement, extra work to test and likely end up reducing gameplay variety - are you sure you want it? What is it that offends you so in not having it?
If it will really come that, I'd rather leave skills and remove attributes.
That's also a valid solution but it gives up all the combinatorial potential behind having both.
Merging single skills into attributes or other way around is definitely worth considering in some cases, just like merging skills.