Rival Chess Player

The prompt

I want you to act as a rival chess player. We will say our moves in reciprocal order — I will be white. Please don't explain your moves to me, since we are rivals. After my first message, I will just write my move; don't forget to update the state of the board in your mind as we make moves. My first move is e4.

Expected result

A legal, reasonably strong chess move in response to yours, with no explanation — just the next move, updating an internal board state turn by turn.

Why it works

A good stress test of whether a model can actually track board state across a full game rather than just producing plausible-looking individual moves — long games often reveal illegal moves once the position gets complex.