Claude Code skill turns Lichess games into narrated chess videos for about $15

A new Claude Code skill can take a vague request like "analyze my last lichess game" and return a narrated video breakdown of the match, according to a Show HN post that drew 71 points and 51 comments. The tool pairs Anthropic's Claude with Stockfish, the open-source chess engine, to generate spoken commentary rather than a raw engine evaluation. It grew out of two earlier experiments by the same developer: whether Claude could read a chessboard through vision alone instead of PGN notation, and whether the model plus Stockfish could explain a game in plain language.

Both experiments reportedly worked. The developer then added a layer that captures his own live audio notes during play and feeds them into the analysis, so the finished video reflects not just what the engine prefers but what the player was thinking at the time. That distinction is what the author calls the most interesting part of the project, since it turns the output into a teaching tool rather than a list of better moves. The system accepts text notes as well as audio.

The tradeoffs are real. A single game takes about an hour to process, and the last analyzed match would have cost roughly $15 at Anthropic's API rates, based on session logs the developer shared. Token consumption is heavy enough that he warned users to check their quota before running a batch. Output quality is also imperfect, though he framed the result as more memorable than clicking through Stockfish branches on a board.

The post lands as developers push Claude Code, Anthropic's agentic coding tool, into tasks well outside software engineering. Chess analysis is an unusual fit because it requires both structured reasoning over engine lines and natural-language explanation, two things large language models handle unevenly on their own. Offloading the numeric evaluation to Stockfish and leaving the narration to Claude is a pragmatic split that sidesteps the model's known weakness at calculating deep tactics.

Several commenters questioned whether the cost and latency justify the result when free tools already annotate games, while others focused on the audio-notes angle as a way to study decision-making under time pressure. The developer has not published pricing or setup details beyond the token estimate, and it is unclear whether the skill will be packaged for other users or remain a personal project. What to watch is whether anyone builds a cheaper pipeline, perhaps by swapping in a smaller model for commentary or caching engine output across sessions.