That's why in my opinion the best model is:
- attributes that pretty much stay the same and depend on race and advantage/disadvantage system at character creation to simulate character's persona. Attributes are there to represent physical and mental setups, and adjust in-game stats accordingly (health, fatigue, speed, magicka, and so on... are all stats that should be determined by your attributes, not by your skills/perks)
- skills that tell how much you have learned in the activities they govern. These can follow the great TES method (they increase if you often use that skill), but you should not be able to max out every skill in the game (why not have them capped at a max level depending on how high your attributes are?
For example, if my Personality is 50, my dialogue skills should be capped at something like 75. This would also mean I can not learn the best tricks aka the best perks related to dialogue skills
). Unlike attributes, skills should determine the outcome of the activities I do in-game (i.e. health is not an activity, that's why it should be governed by attributes ... same for fatigue and so on; on the other hand, crafting outcomes should certainly depend on a skill, not on an attribute).- on top of that, perks are the tricks that let you specialize your style of play in a particular skill (it's good the way you select them in Skyrim, BUT there should be a lot more perks and specializations AND there shouldn't be "increase X by Y%" AT ALL (they're garbage because they're skill-redundant)).
EDIT: Of course, all of the above would work wonderfully and increase replayability dramatically, but it is essential that the game contains a lot of situations where my attributes, skills and perks can be measured against (a dialogue skill should have meaningful impact in the game through dialogue choices; perks should allow to specialize in either admiring OR intimidating, for example).

). They don't break either (well, not anymore