Sphinx

About

A comprehension checkpoint so you still understand the code your agent wrote

Sphinx pauses `git push` on the branches you choose and opens a local questionnaire. The questions come from a reviewer model that never saw your chat with the agent: it reads the committed change through read-only Git tools and asks about behavior, failure modes and invariants, not file names. The check is open book, you can read the code while you answer. Small changes get one question, large ones up to three at the default depth, and questions are often prepared in the background after you commit. Reviewer options are a signed-in Claude Code or Codex CLI (uses that plan), OpenRouter, or any OpenAI-compatible endpoint. Your answers never reach the coding agent, .env files and keys are excluded by default, and every skip or `sphinx yolo` toggle is logged. It is practice, not a security gate: anyone can bypass a hook. Hit the Hacker News front page in October 2026. Repo: https://github.com/parthh01/sphinx.

Features

• Blocks pushes to protected branches until you pass a short check
• Questions written by a separate reviewer that only sees the diff
• Wrong answers get an explanation and a fresh scenario
• Works with Claude Code, Codex, Cursor, OpenCode, Pi or plain Git
• Reviewer via Claude Code or Codex plan, OpenRouter or any OpenAI-compatible API
• Skips and yolo mode are logged with a reason

Install