Sphinx
Git hook that quizzes you on agent-written code before pushing.
- AI tool
- Free
Pricing
Free and open source; reviewer calls use your own model plan or API key. License: MIT.
Overview
Sphinx is an open-source Git hook that checks you understand a change before it reaches a protected branch. When you push, a separate reviewer model reads the commit using Git tools that cannot write and asks you to predict how the code behaves, then explains anything you get wrong and offers a fresh case to reason through. Questions are about how the code behaves, how it fails and what must stay true, and you may consult the code while answering. It supports Cursor, Pi, Codex, OpenCode, Claude Code and plain git pushes, and the reviewer can run on an existing Codex or Claude Code subscription, an OpenRouter key or another endpoint that speaks the OpenAI API. Security note: it sets a global Git hooks path for every repository on your computer and sends changed code to the review model's provider, so set spending limits on paid providers.
Key features
- Holds matching pushes until you pass the check, pause it or skip it
- Separate reviewer model that does not see the coding chat
- Open-book questions on behavior and failure modes
- Supports Cursor, Pi, Codex, OpenCode, Claude Code and plain git
- Per-repository policy and ignore files
Use cases
Alternatives
Last updated 2026-10-19