PRcheck: Prove a human understood the AI code before it merges

Hey

I built PRcheck because of one question: when half your diff is AI-written, did anyone actually understand it before it hit main?

AI-generated code passes tests and looks fine in review. The problem shows up later, during an incident or a refactor, when nobody can explain why the code does what it does.

How it works

A PR opens, and PRcheck posts 1–2 questions like “what breaks if this is wrong?”
The author replies with /explain …
The answer is checked against the diff, and the GitHub check goes green
The “why” is saved to a searchable decision log

Who it’s for: teams merging a lot of AI-assisted code who care more about “can someone explain this?” than about coverage numbers.

What it deliberately doesn’t do

No scores or leaderboards. It measures pull requests, not engineers.
It never blocks a merge because PRcheck itself broke.
It stays quiet on drafts, bots, docs-only and tiny diffs.

v1 was a post-merge quiz. Teams told me that was the wrong moment, so I rebuilt it around the PR itself.

It’s early, and I’d love your feedback on:

Would you make this a required check, or keep it advisory?
What would make you trust a green check, rather than suspect someone pasted a summary?
Where would this annoy your team?

Happy to set it up on a repo for anyone who wants to try it. Just comment below.

Hey