Show HN: Sphinx – A hook that quizzes you on your AI's code before shipping
- ID
- 36794
- Status
- summarized
- Published
- 18 Oct 2026, 12:33 PM
- Fetched
- 19 Oct 2026, 1:57 AM
- Provider
- Hacker News
- Category
- dev-community
- Original URL
- https://sphinx-six-theta.vercel.app/
- Source URL
- https://hnrss.org/best
Summary
- Score
- 6.8
- Created
- 19 Oct 2026, 1:57 AM
- Tags
- Audience
- developersvibe_codersai_agent_users
What happened
Sphinx is a free/FOSS Git hook (npm package sphinx-understand) that pauses a push on branches you protect and has a separate reviewer model read the diff, then quizzes you on what the change does when things go wrong — e.g. whether a retried Stripe webhook double-charges. It requires Node 22.19+ on macOS, Linux, or WSL, runs on minimal tokens via your existing agent, OpenRouter, or any OpenAI-compatible endpoint, and claims compatibility with Codex, Claude Code, Cursor, OpenCode, Pi, and plain Git; `sphinx yolo` bypasses a check and logs the skip. The Show HN thread drew 211 points and 73 comments.
Why it matters
If you regularly ship agent-generated diffs, this is a concrete gate you can install in one command and point at your own model endpoint, and the logged `sphinx yolo` skips give you a real number for how often you'd bypass your own review. The design decision worth copying even if you don't install it: the reviewer never sees your chat history and your answers never go back to the coding agent, so the check tests your model of the code rather than the agent's explanation of it. Note the stated scope — it only pauses branches you configure, and .env files and keys are claimed never to be opened, which you should verify before trusting it with a real repo.
Discussion angle
Does an understanding quiz actually catch bad agent code, or does it become a ritual you click through — and what would you measure (skip rate, wrong-answer rate, escaped bugs) in a two-week trial on one protected branch?