Self-Promotion
Show off what you're working on
Start new thread

Free drift scanner for vibecoded repos

by•

I've been vibecoding all year, and the thing that kept biting me wasn't bugs — it was drift: the gap between code that works and code that's actually production-ready.

So I built DriftDetector:

Free, no signup, ~a minute:

  • Paste any GitHub repo URL (public or private)

  • It scans nine dimensions of production readiness — security, accessibility, reliability, design consistency, AI code governance, and more

  • You get one Production Drift Ratio (0–1, Low to Severe), exact file/line locations, which commit opened each gap, and time-to-fix estimates

It's deterministic — rule-based, zero LLM calls — so the score is consistent run to run, not a vibe. Private repos are scanned in memory: never written to disk, no clone, score, or report kept afterward.


Two asks:

  1. Run it on one of your repos — did the report surface anything you actually care about?

  2. If you use AI coding tools daily, what data would be most useful? What's missing?

Honest feedback welcome, including "this is solving a non-problem."

Cheers,
Jonathan, ReWeaver AI ()

12 views

Add a comment

Replies

Best

Haven't pointed it at a repo yet, but the commit that opened the gap is the useful half and you already have it. The single 0 to 1 ratio is what I'd push on. Nine dimensions collapsed into one number means a repo with one severe security hole and eight clean dimensions scores like one that's mildly sloppy everywhere, and those are completely different Tuesdays. Put the worst dimension next to the ratio and I'd trust the number a lot more.

 - This is exactly the failure mode I worry about with any single score, so thanks for pushing on it. Two partial counterpoints: the ratio is weighted by estimated fix hours, so one severe security hole carries a lot more weight than mild sloppiness spread thin — the two Tuesdays don't score identically. And the report tracks drift per commit, so you can see whether the number came from one bad merge or slow accumulation, which is half of telling those Tuesdays apart, I guess.

You're right that the headline number alone can't tell you which dimension is bleeding (we also show dimension distribution in the report - possibly addressing some of this - not sure), and making someone dig for it defeats the point of a headline. Worst dimension next to the ratio is a cheap, high-trust fix — totally adding it.

If you get a chance to point it at a repo, I'd love to hear how it holds up.

 - shipping today :)