The player's first encounter with any enemy is a blind fight, without being able to tell what their attack is and what its effects will be. (I should've said surprise attacks instead of abilities initially.) In my case, I lost to the wind goblin after not being able to tell how its attack worked. It's fine when their difficulty comes from an ability they have, like the big tree, but when it's something like the three mermaids a player has to
A) juggle the enemy's non-indicative, unintuitive attacks with their own positioning and field obstacles,
B) remember how they operate as enemies, and
C) lose however much real-world time it took them to get to that point.
My point is, it's possible to lose time to a UX limitation and this gets frustrating quick. My suggestion is a monster encyclopedia or codex, possibly designed similar to the spellbook, that updates with information about the enemy and their abilities/attacks/etc upon encounter. Something an attentive player can read and use to formulate a strategy, maybe with some flavor text or worldbuilding (something that, like the music, I bet you already have planned out but haven't implemented yet), but does not directly tell the player how to win.
On that note, it's hard to tell how to play the game at first. It's simple enough, at this point at least, that a single image on the game page with the energy and movement icons circled and explained should be enough. I don't think a mid-run checkpoint system is necessary; it would detract from the roguelite feel, and giving the player more knowledge should alleviate the need for one.