I'm not sure what I'm not understanding, from the examples I've seen it seems a die rolling exercise.
On that note, I could also describe combat as a die rolling exercise.
If I have to tell the players they need X number of successes...
This is actually not mandatory according to the Rules Compendium.
...(snip)... doing some skills checks at that point I'm not narrating anything more than some rules to surpass something, a minigame that doesn't come from the actions of the players, ...(snip)
You do not do some skills without framing a scene and without any player action declaration.
...(snip)... the DM determines what they can do and they roleplay the consequences.
I'm also not sure what you mean when you say the DM determines what they (the PCs) can do - the example provided in the Rules Compendium reflects on the player's deciding on the skills utilised. And nothing stops you from running a skill challenge without limitations on skill use. In fact the Rule Compendium reflects on Secondary Skills.
If you explain them the rules and let them choose the possible actions ...(snip)...
You're supposed to let the players make their own declarations.
...(snip)... the minigame will be expanded to what actions are valid or not for the challenge ...(snip)...
So your concern is that you cannot say yes or tick a success to an ingenious idea provided by the PCs which does not require a skill check?
The first being that the skill or abilities of an antagonist are not taken into consideration.
Antagonists do not always come into play. i.e. weather, elements, nature (mountain, desert), unorganised library...etc
Perhaps overall level is but the actual DC's are set independent of the actual stats, skills, etc. of an antagonist and the fiction is created independent of such as well.
That could easily be remedied with passive DC scores. In 5e you only roll if the result of the action declaration is uncertain so you're not going to set up easy DCs for a higher level party as the results of the action declaration would for the most part succeed. The challenges would have to be significant or at least appropriate.
But what I don't like even more so is deciding beforehand that X successes are necessary to "succeed" at this task,
Do you have a similar issue with deciding on how many hit points an opponent has, or how many opponents to use?
or for that matter that Y number of failures will "fail" the task. It's letting the mechanical process dictate parameters of the fiction as opposed to letting the fiction inform the mechanical processes.
I find this akin to deciding the difficulty of the encounter. You decide in advance whether the combat encounter will be easy, moderate, hard or deadly.
IMO it doesn't take into consideration that the fiction being created doesn't necessarily lead to a resolution of the situation since successes can (and should) be generated for actions that help... what if all you get are X successes of actions that, while helping, do not ultimately do much or anything to resolve the situation... or a more likely issue what if that final success is not a closure type action? This is where I really feel SC's fall down.
I can see how this might become an issue, HOWEVER I can see how one survival check for a 10 day journey might be an issue for some DMs and I can also see how 10 survival checks for a 10 day journey could be called out as a die rolling exercise. As a middle ground I prefer the SC mechanic which enforces scene framing, player action declaration and a shaping of the fiction as a result of the ongoing successes or failures which produces at least a decent narrative in my view.
EDIT: [MENTION=48965]Imaro[/MENTION] nothing stops you (as DM) from narrating the success or failure of the overall SC once the required successes or failures have been reached - which leads to narrative closure.