Horrible solution. Perfect example of how to kill performance by making code shorter. Solutions here are expected to have O(n) and compete only in number of passes through the list, but this one manages to get to O(n^2).

There is an issue for original kata and you better not inherit it. There really should be explicit tests with repeating minimum in the array. For example [3, 4, 3, 5] should produce 6, not 7. I guess, single fixed test is ok.

The solution is incorrect. It fails if the minimum element repeats anywhere in the array.

Ants NEVER collide more than 2 at once in one place. In this case particularly the middle one first collides with the right one and then (after changing direction) with the left one:
(time goes from top to bottom)

.R..R..L.
..R..RL.. <- first collision
...R.LR..
....!..R. <- second collision
...L.R..R

There is a harder version of this kata. https://www.codewars.com/kata/65e4b8877e8acd0dbae9a14f

There is wider kata about computing regular expressions for divisibility tests. https://www.codewars.com/kata/5993c1d917bc97d05d000068

Instead of having single @Test with loop inside (it is actually many tests!) use @RepeatedTest instead.

Added to description and tweaked slightly some tests

Good point! Done

Many stateless classes have their methods non-static despite of possibility to have them static. As there is no reason to be non-static there is no reason to force methods to be static. And yes, there will be more difficult version of the kata featuring infinite iterators.

I know all of this, I plan to add more difficult version

Disastrous performance. How could this be a top voted solution?

Why is it so?