Chess Postmortem Skills turns a lichess game into a narrated video review with Claude Code and Stockfish: Reveneau

Chess Postmortem Skills turns a lichess game into a narrated video review with Claude Code and Stockfish

Chess Postmortem Skills is a new open-source Claude Code project by brumar that takes a lichess game link and a short audio note, and returns an annotated PGN, an interactive HTML board and a narrated video of the whole game with Stockfish-checked commentary. The Show HN post reached 71 points and the repository has 51 stars two days after creation.

Why it mattersA chess player can now sit through their own game the way a coach would review it with them, without opening a Stockfish window or clicking through variations, and the same shape works for any task where an expensive tool and a language model can be pointed at your own recording.

A chess player who wants to know why they lost has two options today, and neither is enjoyable. Clicking through Stockfish branches is fast and cold. A human coach costs money and time. Both leave the player alone with a bar of numbers and no story about the game they just played.

Chess Postmortem Skills, an open-source project released this week by a developer who goes by brumar, adds a third option. The Show HN thread reached 71 points on Hacker News, and the GitHub repository, created on 25 September, has 51 stars and one fork.

What it does

The project is a set of four Claude Code skills, released under MIT with two vendored components under their own licences. The main skill takes a lichess game link and an optional audio note from the player describing what they were thinking during the game. Stockfish runs a sweep of every move to mark the mistakes. Claude Code then reads the player's own notes, matches them to the moves, and writes plain-language commentary that a subagent verifies against the engine before it is included. The output is an annotated PGN file and a standalone HTML analysis board.

A second skill turns the same commentary into a narrated video with a board animation, evaluation gauge, text-to-speech voice, and burned-in subtitles. Two smaller skills cover single-position analysis from a FEN string or screenshot, and playing against Claude through PGN files with a blunder-check subagent.

What it costs

The author writes on Hacker News that his last analysed game cost around 15 dollars at API prices and about an hour of processing. The setup on the player's own machine needs Python with python-chess, cairosvg, pillow and piper-tts, plus Stockfish, FFmpeg and a piper voice for the video. Whisper is optional, for transcribing the audio note.

Why the shape matters beyond chess

The interesting part is the shape of the workflow. An amateur has recorded their own thinking about a decision. An expensive specialist tool holds the ground truth. A language model reads both and writes the annotation that ties one to the other. That shape covers a pitch deck rehearsal against sales data, a code review against static analysis output, and a home renovation walk-through against a building code. Doing that alignment step by hand is slow enough that most people skip it, and the result is that the specialist's output stays with the specialist. A model that does the alignment turns an expensive lookup into something a beginner can sit through.

Source

Primary source: brumar/chess-postmortem-skills on GitHub. Discussion: Show HN thread 49857528.

This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.