Sphinx quizzes you on your AI agent's code before a git push goes through

Sphinx is a Git hook, installed through npm, that pauses a push on branches you protect and asks you a question about what the change actually does, usually how it behaves when something fails. A separate reviewer model that never saw your chat with the coding agent reads the diff and writes the question. Get it wrong and it explains why and tries a new scenario; get it right and the push continues. It works with Codex, Claude Code, Cursor, OpenCode, Pi or plain Git, and can run on your existing agent, OpenRouter or any OpenAI-compatible endpoint. The makers say your answers never reach the coding agent and .env files and keys are never opened, and a "sphinx yolo" command lets pushes through while logging every skip. It is a new Show HN project and has not been independently tested. On Hacker News (241 points, 84 comments), opinion is split. Five commenters say they built something similar: two say quizzing themselves on a change helped their reviews (one adds it does not catch bad design choices), and one says they got frustrated and stopped using it. At least seven dismiss the idea, arguing that people who don't care will skip it, normal code review already covers this, or the real fix is to write or properly learn the code yourself. One reply points out that a lazy user can paste the question into the agent, and another commenter suggests gating pull requests instead, so nobody asks for a colleague's time on code they don't understand. No consensus. Sorted take: The idea is sound for small teams shipping agent-written code; try it on one repo where mistakes cost money, like billing, rather than forcing it everywhere.