Claude Code skill turns chess games into narrated postmortems

A developer using the handle brumar released two Claude Code skills on GitHub that turn a recorded chess game into a narrated postmortem, checking every explanation against the Stockfish engine. The release reached the Hacker News front page Saturday as a Show HN post, drawing 41 points and 28 comments.

The chess-analysis skill runs a full Stockfish sweep of every move. It then sets Claude subagents loose to question the engine the way a person would, asking things like why a move was not played, until each mistake is explained in plain terms rather than a raw evaluation number. It outputs an annotated PGN and a standalone HTML board.

The tool also accepts a player's own think-aloud recording from during the game. Claude transcribes the audio locally with whisper.cpp, then matches each sentence to the move the player was deciding at that moment, using the clock times in the PGN. That lets the postmortem check the player's actual reasoning rather than a guessed one.

A second skill, chess-video, turns the analysis into a narrated video with a board, move arrows, an evaluation gauge and burned-in subtitles, built with the Piper text-to-speech engine and rendered with ffmpeg. A third, chess-play, lets someone play a game against Claude directly, with the board rendered to an image and a separate subagent checking for blunders.

Nobody outside the Hacker News thread has reviewed the tool yet. The developer's own README says a full run, from Stockfish sweep to finished video, takes about an hour of processing. It also warns that Claude can still get a detail wrong, even though a verifier pass re-checks each claim against the engine.