The Sphinx project page does not give a publication time. It says the tool is free and open source. It is a Git hook for agent-assisted development: before the agent’s code ships, it asks a question about what that code actually does. You push, or the agent does. Git pauses on the branches you protect. A reviewer that never saw the chat reads the diff and asks what the change does when things go wrong. A wrong answer gets an explanation, then another scenario. A right answer lets the push continue.
[1]
The page shows one exchange and labels it illustrative. Real questions, it says, come from your diff. That exchange is not a log of a real push, and it is not a tested question bank. The page also says it is built to run on minimal tokens, using your existing coding agent, OpenRouter, or any OpenAI-compatible endpoint. “Minimal” is not given as a token count.
[1]The page names Codex, Claude Code, Cursor, OpenCode, and Pi, and also plain Git. Answers never reach the coding agent. Environment files and keys are never opened. If you are in a hurry, sphinx yolo lets the push through, and every skip is logged. The skip leaves a log. The page does not say where that log lives, or who can read it.
[1]The stated runtime is Node 22.19 or newer, on macOS, Linux, or WSL. The page gives no rate for whether the questions are answered correctly, no count of blocked pushes, and no independent retest. After install it asks which model should review. How the model is chosen and how questions are written are left to the docs. This draft does not expand them.
[1]