Darren as usual provides the concise and correct answer, I can now answer my own thoughts on this matter...
We try to frame this as the "Seven Day Roguelike Challenge" rather than "Contest" to underline that it isn't something to compete at. Despite this, we've faced slop submissions before AI was a thing. Quite a few seem to think great fame occurs by entering the challenge. Fortunately they mostly submit the first few hours making it simpler to weed, but this is why we have any submission criterion at all.
Speaking personally, I don't mind if your @ on a map is boring and derivative if you "made" it.
At the end of the day we want to encourage intrinsic rewards. As Darren says, your own satisfaction is the goal.
lsdcomputer has nailed the spirit. They want the challenge of making their own roguelike. For them, vibe coding defeats the entire point - they might as well just download nethack and call it a day. Another coder might feel using libtcod is too much a crutch; so for them they had better eschew that. But that doesn't mean lsdcomputer should avoid libtcod!
But you do ask the interesting question: what part is the part that earns "respect". Despite the risk that I'm now externalizing rewards, I'll try to answer that from my personal viewpoint.
The reason I advocate allowing libraries, allowing using existing assets, and not having a "theme", is because I'm not seeking "game design" or "lines of code" or "pretty jpegs". I'm seeking Art.
The very first Seven Day Roguelike, Dungeon Monkey, was created as just that: Performative Art.
And I think that is the part of the process that I most respect. The artistic process. Art really has two stages. One part is your internal vision, what is living in yourself you wish to express. The second is the external embodiment of that vision. This is the artifact you present at the end of the seven days, for us to fail to interpret to your satisfaction. The artistic process is trying to build that artifact - this has an "Embodied Intelligence" component. When people say that a stone talks to them and tells them what should be carved from it; I don't believe there is actually some spirit in the stone talking. But I do believe that is the correct way to describe the situation.
Over the seven days as you build a game; the game will talk to you. As you talk to it. So having your hands on the game; writing dialog text; colouring pixels; typing semicolons; all that is often an essential part of the process of the game coming to exist. That often forms the equivalent of the brush strokes of the painter, which lead to the reflection that allows the many micro decisions required to take shape to bring what is in your mind into the world. Because, if you took a brain scan of me before the 7DRL starts and assume the final game is sitting latent in that space; it isn't; it does result from a dialogue rather than a dictation.
Unfortunately I don't recall the name, but there was one 7DRL I played that had quite a few simple but clever reframings of the genre. I source dove it out of curiousity and was a quite surprised to see they author had "merely" walked through the libtcod tutorial. But in the act of doing so they had stamped it with their art and their story.
Fortunately ASCII is respected in Roguelikes; so there is no need to consider image generation. I'd strongly recommend not fighting with prompts over what your creatures should look like when a simple glyph would suffice - spend that energy in the creature description.
As for vibe coding; I'd believe you can make a great roguelike vibe-coding. I don't believe it would be "easier" to do so, however.
So, no, there are no restrictions.