chess-digest reviews your recent chess.com games. It runs Stockfish on every move, asks Claude which mistakes you keep repeating, and writes a short report with board diagrams and links to practice material.
19. Qxb6+ Kb8 20. Qxb7#Every report is one HTML file you can open anywhere. It focuses on patterns, not on single bad moves.
Three to five patterns ranked by how much they cost you, each with a way to fix it.
Boards from your games with your move in red and the engine's move in green.
Lichess puzzle themes, Lichess practice and Wikipedia pages picked for each problem.
Record, accuracy, blunders per game, openings and mistakes by game phase.
Stockfish decides what was a mistake. Claude only groups them and explains them.
Games come from the chess.com Published-Data API. You don't need to log in.
Stockfish checks every position. A drop of 20 points in winning chances is a mistake, 30 is a blunder.
Claude gets the worst mistakes in one request and can only pick themes and positions that exist.
You get one HTML file. Games from earlier reports are skipped next time.
You need Docker Desktop, git and an Anthropic API key.
$ git clone https://github.com/VladKhomich/chess-digest.git $ cd chess-digest $ cp .env.example .env # key + username $ docker compose build $ docker compose run --rm digest --games 30
# once, in $PROFILE . "C:\path\to\chess-digest\chess.ps1" > chess analyse -g 10 # last 10 games > chess analyse --week # last 7 days > chess open # latest report
Reports land in out/. Add --all to include games from earlier reports, or --no-ai for a free engine-only report.
Fork the repo and turn on GitHub Pages for the docs folder of main. Then run chess analyse --week --publish.
It commits only docs/digests/ and pushes. Published pages hide your opponents' usernames and ask search engines not to index them.
Stockfish runs on your machine. Each report makes one Claude Sonnet call.
That's usually a few cents, and a week of games rarely goes past 25 cents.
54 games, 187 flagged errors and five patterns worth fixing.