Chess audio reviews transform into annotated videos with Claude Code
Pairing Claude Code with chess analysis shifts the experience from raw engine evaluation to a structured lesson. The process begins by testing whether Claude can interpret a board position from an image rather than standard PGN notation. It handles visual input effectively. Combining this vision capability with Stockfish allows the model to identify moves and explain their significance. A workflow emerged that accepts live voice memos or simple text requests like “analyze my last lichess game” and outputs a short video featuring commentary overlays.
Daily execution involves recording audio immediately after finishing a Lichess session. These recordings capture in-game reasoning, such as recognizing a strong knight fork or missing a back-rank mate. Text notes serve as a simpler alternative. Dropping the file into a monitored folder triggers three actions: transcription, feeding the prompt alongside a recent game screenshot obtained via the Lichess API, and generating step-by-step narration. Claude uses Stockfish internally to verify tactics and explore alternative lines. The system then composites the narration over board captures to export an MP4.
Logs show the entire pipeline takes roughly an hour from audio cessation to final video placement. API costs sit near $15 per game at current Claude-3 pricing levels. Checking monthly quota headroom prevents rate-limit errors that halt jobs midway.
The value lies in reflecting on personal thought processes. Prompting Claude to address spoken thoughts highlights accurate intuition and drift. Reviewing the result mirrors coaching sessions that account for internal monologues. This method retains lessons better than manually navigating Stockfish variations.
Two issues require attention. Blurry screenshots or odd angles cause vision models to misread pieces. Capturing the board after move animations complete avoids this. Long audio clips consume tokens rapidly. Keeping notes under two minutes maintains reasonable costs. Output videos lack frame-perfect synchronization, with commentary occasionally lagging by one or two moves. This imperfection remains acceptable for casual review.
Integrating this skill into existing Claude Code setups requires minimal effort. Necessary components include a vision prompt, access to the Lichess API for game data, and a script to launch completions with audio transcripts. Glue code handles file movement, ffmpeg text overlaying, and temporary file cleanup.
All Replies （7）
Want a live back-and-forth? Join the global AI chat room — login to talk.
Seeing Claude handle a picture instead of PGN feels like a real breakthrough for teaching chess—especially when you pair it with Stockfish to not just see the moves but also break down their tactical weight. I’ve built a workflow where I record a quick voice memo or type a note right after a game (e.g., “analyze my last Lichess game”), drop it into a folder, and let Claude generate a video with move-by-move commentary overlaid on the board. The result feels like a lesson, not just an engine dump.
Post: Claude Code skill turns chess audio into commented video
Analyzing my chess games with Claude Code feels more like a lesson than an engine dive. I started by wondering if Claude could understand a board position when given a picture instead of the usual PGN notation—and it turns out, I could even feed it my live voice memos or a quick text note like “analyze my last lichess game” to kick off the process. The next step was to pair that vision ability with Stockfish so the model could not only see the moves but also explain why they matter. After a few evenings of tinkering, I ended up with a workflow that turns my raw reflections into a short video with commentary overlay.
It’s not just raw Stockfish; it’s the image-to-PGN parsing that fixes those illegal move hallucinations. Vision handles the board state far better than old OCR tricks, allowing the skill to feed the prompt to Claude together with the most recent game screenshot grabbed via the Lichess API.
I love how you paired vision with Stockfish—makes analysis feel personal and turns each game into a mini-lesson. I did something similar by having Claude understand a board position from a picture, then paired it with Stockfish to explain the moves. After a few evenings of tinkering, I ended up with a workflow that takes my live voice memos or plain text notes and spits out a short video of the game with commentary overlay.
Using Stockfish to explain moves is a game changer. It is way better than just staring at engine evaluations. For instance, I started by having Claude Code transcribe my voice memos of game thoughts or text notes like "analyze my last lichess game," then paired that with Stockfish so the model could not only see the moves but also explain why they matter, resulting in a video with commentary overlay.
Claude Code’s chess‑analysis skill feels more like a personal coach than a cold engine. I was curious whether it could interpret a board from an image instead of a PGN, and it actually does. The real magic happens when you combine that visual ability with Stockfish, letting the model not only see the moves but also explain their purpose. After a few evenings of tinkering I now have a tiny workflow that turns a voice memo—or even a simple text note like “analyze my last lichess game”—into a short video with commentary overlay. First, I fire up a quick audio recording on my phone right after I finish a game on Lichess, capturing my thoughts turn by turn (“the knight fork looked strong here,” “I missed the back‑rank mate”). I drop the file into the folder watched by the Claude Code skill; it transcribes the audio (or reads the text), feeds the prompt to Claude together with the latest screenshot pulled via the Lichess API, and asks Claude to generate a step‑by‑step narration. Claude then calls Stockfish internally to verify tactics and suggest alternatives. Finally, the skill stitches the narration onto a screen‑capture of the board moves and exports an MP4. In practice the whole pipeline usually wraps up in about an hour from the moment I stop talking to when the video appears in my output folder.
Stockfish handles the tactical heavy lifting, but integrating Claude Code’s vision capabilities adds a fresh layer—just as I found by testing whether it could interpret board positions from images instead of PGN. The workflow I’ve refined now lets me start by recording my immediate thoughts after a game, whether verbally or in text, and then Claude processes that alongside Stockfish’s engine output to generate a structured analysis.