How Can AI Explain a Chess Mistake? Claude Code’s Post-Game Skills
Claude Code
Anthropic’s tool for helping people write code.
Stockfish
Software that calculates chess positions and moves.
PGN
A file format for storing chess moves.
What was released
A public GitHub repository called chess-postmortem-skills adds a set of skills to Claude Code, Anthropic’s coding tool. It turns one chess game into several readable artifacts: an annotated PGN, a standalone HTML analysis board, and a narrated video. It also includes separate skills for analyzing one position and playing against Claude.
How it turns a mistake into an explanation
The workflow begins with a Stockfish sweep of every move. It flags mistakes at a stated threshold of at least 0.3 pawns of loss. Investigator subagents then ask the engine why the played move failed. They also question non-obvious moves inside the suggested variations. The workflow renders important positions, builds the annotations, and runs a verifier. One planned pause per game explains a quiet position where the player’s plan drifted, rather than focusing only on one blunder.
A think-aloud recording is optional. When supplied, the project uses whisper.cpp locally, then aligns the transcript with move times from the PGN clock. This lets the review address what the player actually considered. Without a recording, it can still produce an engine-based explanation.
The public example
The author tested the workflow on a personal 15+10 rapid game on Lichess. The author played White in an Open Sicilian against a Najdorf, at about 1700 rating. During the game, the author recorded thoughts in French. Claude received the Lichess game and the audio. The resulting video runs 6:49, with English narration and subtitles. The full run takes about one hour.
Why it matters
A chess engine can show a score and a long variation. Those outputs do not automatically explain the player’s fear, plan, or question. This project tries to move the center of review from finding the best move to explaining why the played move failed. A recording makes that explanation more personal. The system can test a real thought instead of inventing one after the game.
That is a useful design idea, but it is not the same as proving that AI gives reliable coaching. The repository describes an engineering workflow and a public example. It does not present an independent benchmark across many players.
What is confirmed, and what is not
The repository includes the skills, example files, requirements, and a finished video. It says claims are checked against Stockfish and a verifier pass. It also warns that AI can still be wrong. The statement that errors are rare is the author’s experience, not a published accuracy rate.
On Hacker News, the project received 73 points and 53 comments. That measures community attention. It does not confirm the project’s correctness or usefulness.
What to watch next
The important tests are practical. Does the explanation stay clear for different playing levels? Does it work without a think-aloud recording? Can another player’s game reproduce the same quality? Is an approximately one-hour run worth the setup? Those answers will show whether this is a personal workflow or a broadly useful coaching tool.
Sources: the GitHub repository, the Hacker News discussion.
An AI Tool That Explains Chess Mistakes
📰 Full story: How Can AI Explain a Chess Mistake? Claude Code’s Post-Game Skills
A public project asks Claude Code to turn chess-engine numbers into understandable lessons.
Claude Code
Anthropic’s tool for helping with code.
Stockfish
A program that checks chess moves.
PGN
A file that stores a chess game.
💡 The gist
- A GitHub project helps Claude Code explain chess mistakes.
- Stockfish checks moves, while Claude asks why they failed.
- Hacker News showed interest, but attention is not proof.
Anthropic is the company behind Claude. Claude Code is its coding tool. This project adds chess skills to that tool.
A player gives it a game. The game can include a Lichess link and a PGN. PGN is a file that stores chess moves. The project can make an annotated game file, a web board, and a narrated video.
First, Stockfish checks every move. It finds moves that lose value. Then helper agents ask simple questions. Why was the played move bad? What happens after the natural reply? The system checks important positions. A final verifier checks the written analysis again.
The project can also use a voice recording. The player talks while thinking. whisper.cpp changes the recording into text. The workflow matches those words with the moves. This matters because the review can answer the player’s real idea. It does not need to guess what the player thought.
The author tested the project on one personal game. It was a 15+10 rapid game on Lichess. The author played White in an Open Sicilian. The video lasts 6:49. It has English narration and subtitles. The complete run takes about one hour.
This approach matters for one reason. Engine numbers show that a move lost value. They do not always explain the human mistake. A clear reason may teach more than a score. It can help a player remember a pattern next time. That is the project’s promise, not a measured result.
Still, the public example is mainly one author’s workflow. It is not a large accuracy study. The author also says AI can make mistakes.
The project reached 73 points and 53 comments on Hacker News. These numbers show attention from that community. They do not prove that every explanation is correct.
The next test is wider use. Other players should try different levels and games. Reviews should also be tested without recordings. That will show whether the tool is useful beyond its creator’s example.
💬 HN comment digest, easy: Can AI be a chess teacher?
This tool uses Stockfish to calculate chess positions and another AI to explain the results. Some commenters see a useful teacher’s aid; others worry about mistakes and false confidence.
- Stockfish calculates moves, and the LLM explains the reasons. Some commenters say the engine and the agent can work together, but the author has not proved that this improves learning. Reports that raw LLM chess is unreliable are personal reports, not a full benchmark.
- The idea is not completely new. Chess engines have analyzed games for years, and existing services may already add AI-style explanations.
- The learning benefit is unclear. Talking through your own thinking with AI may help, but a smooth explanation can also make someone feel they understand more than they really do. Simply browsing engine variations can be passive too.
- A user who tried Claude and Stockfish for several hours reported that the analysis was not useful. Other users reported illegal moves and invented pieces. These are individual experiences, not a complete performance test.
- Quiet positions and strategic plans are especially hard. The author also admits the output can be difficult for a weaker player to check, so the tool is safer as a helper checked against Stockfish or a stronger player than as a standalone coach.
initial digest at 53 comments (revision 1). We fetched 51 comments and sampled 51 across the thread. These are HN users’ reports, not independently verified facts.
A Computer Helps Find a Chess Mistake
📰 Full story: How Can AI Explain a Chess Mistake? Claude Code’s Post-Game Skills
A computer looks at chess moves and helps explain what went wrong.
Claude Code
A tool that helps people write computer code.
Stockfish
A computer program that checks chess moves.
Hacker News
A website where people share technology news.
Claude Code is a tool from Anthropic, the company that makes Claude.
It can help look at a chess game.
Stockfish is a chess program that checks moves.
It can find a move that lost.
Claude Code then asks why the move failed.
It checks the board and writes an explanation.
A player can also record their thoughts.
The tool matches those thoughts with the moves.
Then the player can see what they believed.
They can see what the chess checker found.
The project also makes a board and a talking video.
The creator says the computer can still make mistakes.
Hacker News is a technology link-sharing site.
This project got 73 points and 53 comments there.
Those numbers show attention.
They do not show perfect answers.
💬 HN comment digest, for a 5-year-old: A chess AI that talks
One computer checks the chess moves, and another AI explains them. It might help, but it has not been shown to be a real teacher yet.
- The calculator looks for a good move, and the talking AI explains why. Some people like putting the two together, but the author has not proved that it helps people learn.
- Similar chess calculators have existed for a long time, so the whole idea is not new. A pretty AI explanation can also make someone feel they understand when they do not. Looking at calculator answers without thinking can be passive.
- One user said that several hours of trying Claude with Stockfish did not produce useful analysis. Other users said the AI sometimes gives strange moves or talks about pieces that are not there. These are user reports.
- Quiet chess positions are especially hard to explain. So it is best to check the AI with the chess calculator or with a stronger player.
initial digest at 53 comments (revision 1). We fetched 51 comments and sampled 51 across the thread. These are HN users’ reports, not independently verified facts.
💬 HN comment digest: Can Stockfish plus an LLM teach chess?
The discussion centers on a Claude Code skill that uses Stockfish as the calculation layer and an LLM as the explanatory layer. Commenters see a possible teaching aid, but disagree about its novelty, reliability, and ability to build deep understanding.
initial digest at 53 comments (revision 1). We fetched 51 comments and sampled 51 across the thread. These are HN users’ reports, not independently verified facts.