FromPosition matches with FRC castling rights are accepted and correctly handled by the Lichess GUI, but cause issues for lichess-bot

[Naphthalin]

Describe the bug
lichess-bot bots can't handle FRC FromPosition games, will receive wrong castling rights from the server, and error out after the opponent castles "illegally". https://lichess.org/rQA6AjkX as an example.

To Reproduce
Steps to reproduce the behavior:

1. Challenge a lichess-bot who accepts arbitrary FromPosition challenges to a position with FRC castling rights included, like rbqnkrbn/pppppppp/8/8/8/8/PPPPPPPP/RBQ1KRBN w AFaf - 0 1
2. Castle by capturing your own rook
3. Bot errors out because it can't set up the correct position and will time out.

Expected behavior
https://github.com/lichess-bot-devs/lichess-bot/blob/master/lib/lichess_bot.py#L832-L835
This should also set the mode to chess960=True if it's a "From Position" but with FRC castling rights. However, a quick test shows that game.initial_fen seems to have standard castling rights, so this might require some additional logic.

Logs
Throws a uci move when handling the FRC castling:

line 3147, in parse_uci
                                 raise IllegalMoveError(f"illegal uci: {uci!r} in {self.fen()}")
                             chess.IllegalMoveError: illegal uci: 'e8f8' in
                             rbqnkr1n/ppppp1pp/4b3/5p2/2P2P2/2Q5/PP1PP1PP/RB2KRBN b Qq - 2 3
[10/10/25 09:36:18] ERROR    Ignoring illegal move e8f8 on board
started from rbqnkrb

Desktop (please complete the following information):

• OS: [e.g. Windows]
• Python Version: [e.g. 3.11]
• Terminal: [e.g. Command Line, PowerShell]

Additional context
Add any other context about the problem here.

[MarkZH]

Can you post the complete log from lichess_bot_auto_logs from the game that had the error? Among other things, I want to see the messages the lichess sends to start the game.

[Naphthalin]

Slightly different position, where black can castle on the first move:
https://lichess.org/Yts96MAJ
FRCfromposition.log

As far as I can see, the following happens:

• the challenge is received with 'initialFen': 'rnbqnkrb/pppppppp/8/8/8/8/PPPPPPPP/RNB1NKRB w AGag - 0 1'
• the game is started with 'initialFen': 'rnbqnkrb/pppppppp/8/8/8/8/PPPPPPPP/RNB1NKRB w KQkq - 0 1'
• uci sends position fen rnbqnkrb/pppppppp/8/8/8/8/PPPPPPPP/RNB1NKRB w - - 0 1

[MarkZH]

Please post the whole log file. It will help us determine how to fix this. Use gist and post a link if it's long.

[Naphthalin]

While I'm pretty sure the above linked file contains everything relevant, here is the full log file:

https://gist.githubusercontent.com/Naphthalin/a9bbadbc466fc942acfe465ff518282e/raw/ec1df0336d0838882dd2c428d04770c81d94c660/lichess-bot_20251010.log

Thanks for looking into this!

[MarkZH]

Sorry, I didn't see the second link beneath the lichess link.

Once communication with the engine starts, it looks like there's no indication that the position is chess960 except for castling rights with rooks in non-standard positions.

[MarkZH]

Can you try PR #1147 and see if that fixes the problem?

[Naphthalin]

Thanks, this looks good! It now has the side effect that I also need to accept Chess960 challenges in the config.yml (not sure if this intended with the PR). I tested it in https://lichess.org/70VknwB5

What is slightly weird is the FENs in the log file:

• challenge: 'initialFen': 'rbqnbkrn/pppppppp/8/8/8/8/PPPPPPPP/RBQ1BKRN w AGag - 0 1'
• game start: 'initialFen': 'rbqnbkrn/pppppppp/8/8/8/8/PPPPPPPP/RBQ1BKRN w KQkq - 0 1'
• uci: position fen rbqnbkrn/pppppppp/8/8/8/8/PPPPPPPP/RBQ1BKRN w GAga - 0 1
• FEN at the end of the game: 'fen': 'rbqnbrkn/ppp1pppp/8/3p4/2P3P1/4P3/PP1P1P1P/RBQ1BKRN b HQ - 0 3'

Doesn't seem to cause any issues, but might do depending on how strictly the uci protocol and FEN notation is followed.

[MarkZH]

I tested the PR with my own bot that doesn't play chess960 and got errors with your example FEN. So, I think it's necessary for the user to indicate that a bot can play 960 games and otherwise decline challenges with 960 positions.

The last FEN with HQ looks odd, like it's a mix of standard and FRC castling rights notation. That might be worth submitting a bug report to the lichess API repo: https://github.com/lichess-org/api

[MarkZH]

Both notations AGag and KQkq are reasonable--the former so an engine doesn't have to search for the rooks and the latter because it's unambiguous. But, HQ is a mix of these and not even correct. There's no rook on the H-file to castle with. It should be GQ if anything.