Sphinx

Sphinx is a local Git hook by GitHub user parthh01 for people who write code with AI assistants. When you push to a protected branch (main and master by default), it pauses the push and has a separate reviewer model read the committed change, then asks you how the change will behave and where it could fail. The check is open-book. A wrong answer gets an explanation and a new question on the same idea; a pass lets the same push continue. Its tagline is "You can outsource your thinking, but not your understanding." It works with Claude Code, Codex, Cursor, OpenCode and Pi, or with a plain git push and no assistant.

Who it's for:Developers who let AI write code and want to be sure they understand each change. Readers who don't write code won't get much from it.

Before you start: You need Node.js 22.19 or later and Git, on macOS, Linux or WSL; native Windows isn't supported yet. Install with npm install -g sphinx-understand, then run sphinx install. That command changes your global Git settings so pushes in every repository on the machine are checked, and it changes your chosen assistant's settings; sphinx yolo turns checks off for a repository. Questions come from your own model account: a signed-in Codex or Claude Code, an OpenRouter key, or an OpenAI-compatible endpoint, billed separately. There's no default cap on model calls; the author says a typical check makes about 5 to 8 requests, and you can set a limit. The changed code, related files and your answers go to the reviewer model you choose. .env and common credential files are excluded by default, but secrets written into ordinary code are not. The author describes it as practice, not a security gate, and says the reviewer model can be wrong. MIT-licensed.

Based on the project's public documentation. Not tested by AI Inspiration World.