Sphinx offers Git hook for AI-code understanding checks

Sphinx, a free open-source Git hook, appeared on Hacker News on Oct. 18 with 16 points and 2 comments, pitching a check on agent-assisted software changes before deployment.

The product page says Sphinx reviews a diff before a push proceeds; its illustrative case covers 11 files and 8,159 added or deleted lines combined, according to Atlas360 calculations.

The tool is aimed at teams using AI coding systems such as Codex, Claude Code, Cursor, OpenCode and Pi, or plain Git. Its core claim is narrow: before code generated with an agent ships, Sphinx asks the developer a question about what the change does.

The page frames the problem with the line, “Outsource your thinking, not your understanding.” It says real questions are generated from the code diff, while the example push shows +5,012 and -3,147 lines.

In that example, Sphinx asks about a webhook handler that saves a Stripe event ID before calling charge(). The user first answers that a retry charges twice; Sphinx responds that the real failure mode is that a failed charge may be skipped on retry because the ID was already stored.

Sphinx says it can be installed with npm and connected through `sphinx install`, which links Git and coding agents and asks which model should review. The page says it can run through an existing coding agent, OpenRouter or an OpenAI-compatible endpoint.

The project’s own page says answers do not reach the coding agent, and that `.env` files and keys are never opened. It also offers `sphinx yolo` to let pushes through, with skips logged.

Source: product page and Hacker News post, Oct. 18, 2026