So, instead of drawing X cards, the player rolls X 20 sided dice:

1:-1 success. Yes, it can reduce another success.

2-12: 0 success

13-19: 1 success

20: 2 successes, and reroll, and add new result (if the plyer rerolls a 1, then it removes a success, if he rolls a 20, then 2 new successes are added, and the dice is rolled again)

You understand how statistically this is quite different from resolving a test based on a set of cards having a fixed number of successes, right? Just checking; if so, then expanding houserules is 100% fine

I do. I did the math, considering the number of cards and successes printed on the cards. However, indeed, no cards have a 2 successes or -1 success printed on them, and the probability for exhausting a deck of success cards and throwing a die or dice is not the same. However, it is then possible to score some "critical" success (like this time I gained 8 golds from some action, after rolling multiple 20s), or failure, and imho, it adds to the fun, and rpg feel, of it.

Sure thing I wasn't debating the idea behind it, just making sure it's clear there's a difference

Hmm. So can I use this system if I use every skill card from all the expansions and the die rolls will have roughly the same odds as if I drew the skill cards for tests per the rules?

I am trying this in my games and I love it, I'll never use the cards again, thank you!

No.

Maybe I will try this, love rolling dice!