Show HN: Sphinx – A hook that quizzes you on your AI's code before shipping

What it claims

Sphinx is a free, open-source Git hook that blocks a push to a protected branch until the developer correctly answers a question about the diff. The questions target failure behaviour. The page's example concerns a Stripe webhook handler that saves the event ID before calling charge(). The follow-up asks what a retry does if charge() fails, and the correct answer is that the customer is never billed. The page says Sphinx is built to run on minimal tokens.

How it works

The reviewer model reads only the diff, not the coding chat. A wrong answer gets an explanation and then a new scenario, so the loop repeats until an answer is right. A sphinx yolo command bypasses the check, and every skip is logged. The page states two data boundaries: the developer's answers never reach the coding agent, and .env files and keys are never opened.

flowchart TD
    A[git push to protected branch] --> B{sphinx yolo?}
    B -- yes --> L[Skip logged, push continues]
    B -- no --> C[Reviewer reads diff only]
    C --> D[Asks failure-case question]
    D --> E{Answer correct?}
    E -- yes --> F[Push continues]
    E -- no --> G[Explanation shown]
    G --> D

How to use it

The tool requires Node 22.19+ and runs on macOS, Linux or WSL. Native Windows is not listed.

npm install -g sphinx-understand
sphinx install
# then push as usual
git push

sphinx install wires Git to the installed coding agents and prompts for the reviewer model. The reviewer can be the existing coding agent, OpenRouter, or any OpenAI-compatible endpoint. Pushes made by an agent are gated the same way as manual ones.

Limits

The page is a landing page. It gives no licence name and no token or cost figures beyond "minimal tokens". It also gives no evidence that the questions are accurate or useful. How protected branches are chosen, and where skip logs are stored, is left to separate docs that the page does not include.