Sphinx: Open-Source Git Hook That Tests Developer Understanding of AI-Generated Code Before Deployment

Sphinx is a free, open-source Git hook designed for agent-assisted development workflows. Its core function is to pause a git push on protected branches and ask the developer a specific question about the code changes (diff) generated by an AI agent. The push only proceeds if the developer answers correctly, ensuring human understanding of the shipped code.

How It Works

The tool operates as a pre-push hook that analyzes the diff rather than the chat history. It uses a reviewer that never saw the chat history to generate questions based on edge cases or logic flows in the new code.

• Workflow: When a developer (or their agent) attempts to push, Sphinx intercepts the action.
• Question Generation: A reviewer reads the diff and asks a question about what the code does in specific scenarios (e.g., error handling, race conditions).
• Verification: The developer must answer correctly. If they fail, they are prompted to try a new scenario. If they succeed, the push continues.
• Bypass: A sphinx yolo command allows bypassing the check, though this action is logged.

Integration and Privacy

Sphinx is designed to be lightweight and privacy-focused, ensuring that the developer's answers do not leak back to the coding agent.

• Compatibility: Works with Codex, Claude Code, Cursor, OpenCode, and Pi, as well as plain Git.
• Model Support: Can use existing coding agent models, OpenRouter, or any OpenAI-compatible endpoint.
• Security: The tool explicitly states that .env files and keys are never opened and developer answers are not sent to the original coding agent.
• Installation: Available via npm (sphinx-understand) and requires Node.js 22.19+.