The title here on HN is "fully working", which isn't true.

Agreed, I wasn't complaining, I was just pointing out a minor inaccuracy in the title. It's very impressive and doesn't need to be oversold.

Castling is pretty important move in chess. You can hardly claim an implementation to be fully working without it (although it wasn't the author claiming that).

I wonder how many characters of code it would take.

The problem with castling is that for a move that is only once done in a chess game we have to keep track of a couple of variables and do a check for every move in the tree. It is rather unpleasant to code and it will increase the program size and decrease the speed (by a bit I mean). May be that is because they ended up not implementing castling.

it's one of three special cases where it isn't moving a piece from and to a square. The other being capturing enpassant, and queening a pawn. Certainly cannot claim to be complete without being able to play a legal move.

...promoting a pawn

...and then there's the 50 move and repetition draws

...and a non-chess player may not realize how vital a knowledge of openings is

but it basically works

Also doesn't detect stalemate.

No, it just sits there.

Does it not report win/loose? I only played one game.