Really cool, love the concept, though it's reliance on trial-and-error really does bring it down somewhat.
The issue is the way the game requires you to choose the correct ability for a given situation before even knowing what obstacles are ahead. The design requires the player to just throw themselves in blind until they encounter a roadblock, then restart and bring the ability that will let them past it. Then they encounter the next roadblock and repeat. And again, and again, and again. This leads to repetition where the player is not learning skills or improving at the game, just trying things until they find what the game wants. This leads to frustration and such.
Much of the initial difficulty comes from the fact that players will generally not have the mere capability to overcome an obstacle when they first encounter it, so will fail through no fault of their own. As opposed to having the capability to overcome an obstacle when they first encounter it and having to retry because they just weren't good enough. I personally gave up on level 3, when it started requiring trickier timing and platforming abilities in combination with the trial and error.
I like how the abilities lend the levels to being more like puzzles, but with no foreknowledge the player is very unlikely to have an appropriate ability the first time it's needed, meaning the solution can only be found through the aforementioned trial and error. So, bad puzzles. Doing something like showing the player an overview of the level before they select abilities (or while they select), would go a long way. This would allow the player to plan ahead and prepare, give them a sense of which abilities might get them through the level, and allow them to tweak their ability set when it doesn't work out until they get a solution that works for them, emphasizing the puzzley nature.
In retrospect I think Solifuge got at a lot of what I'm trying to say in his suggestion #1, only a bit more succinctly.
I did like the game though, possibly more than my critique lets on (gave a pretty good rating for what it's worth).