Show HN: A Claude Code skill to analyze your chess games

TL;DR Highlight

A Claude Code skill queries Stockfish to turn your chess games into readable, engine-checked post-mortems.

Who Should Read

A developer who wants to build verifiable analysis workflows by wiring an engine or external tool into Claude Code skills

Core Mechanics

Claude drives Stockfish itself
- Claude asks the engine questions like "why not the move I played?".
- It keeps asking until the answer makes sense, then writes it up in plain language.
Think-aloud audio lets it refute your actual reasoning
- Audio recorded during the game is transcribed locally with whisper.cpp.
- PGN clock times align each sentence to the move being decided.
Analysis is split into subagent stages
- A Stockfish sweep of every move comes first.
- Parallel investigator subagents then question the engine until each mistake is explained.
Hallucinations are reduced by engine checks and a verifier pass
- Every claim is checked against Stockfish and a verifier pass re-reads the result.
- The author says errors are rare, and you can ask Claude to recheck any odd position.
Four skills cover different jobs

Community Reactions

Objection that LLMs alone are useless at chess
- Commenters said LLMs make illegal moves and hallucinate pieces.
- Others found it interesting that Stockfish backs the LLM, which only adds commentary.
Not novel, some said
- One comment said engine-based automatic analysis has long existed and chess.com likely did it already.
- Others still saw value for learning and curiosity.
Experience: feeding real data into the prompt works
- One commenter built an LLM explainer for a scrabble AI with Monte Carlo data in the prompt, and it hallucinated little.
- Another preferred short, checkable reports over helpers that try to be a whole coach.
Criticism of AI-written SKILL.md and cost
- One comment called letting AI write the prompts lazy.
- One comment questioned a cost of $15 per use.
Question about opening in a short form

How to Apply

Install the skill and run one game today
- Git clone and copy the skills folder into ~/.claude/skills/.
- Tell Claude Code "install the dependencies for the chess skills" to set up requirements.
Copy the pattern of pairing an LLM with a verifier tool
- Make the LLM call a dedicated engine or calculator where it is weak.
- Instruct it to check every claim with that tool before the final answer.
Add a verifier subagent pass
- Run investigator agents in parallel, then add a step that re-reads the results.
- Have it recheck only the suspicious items with the tool.
Pin user context in CLAUDE.md
- Write stable details like reader level into CLAUDE.md once.
- For example: "Write chess analysis for a 1800 player."

Try It

copy & run

# Paste into Claude Code (after installing the skill)

# Install (terminal)
git clone https://github.com/brumar/chess-postmortem-skills
cp -r chess-postmortem-skills/skills/* ~/.claude/skills/

# Type into Claude Code
analyze this game: {{lichess link}}, here is my think-aloud recording: {{mp3 path}}

# Next, try
make a video of this game

Terminology

StockfishAn open-source chess engine. Think of it as a calculator that gives an evaluation and best lines per move.

PGNA text format for chess game records. This project uses its clock times to align audio to moves.

whisper.cppA local implementation of the Whisper speech-to-text model.

Claude Code skillA unit where SKILL.md describes a procedure that Claude reads and follows. The doc is the executable manual.

piper TTSA text-to-speech tool. Used for the video's narration voice.