How I catch AI code mistakes before they break my project

Source: Dev.to

I built a VS Code AI copilot that not only generates code but also catches subtle AI mistakes before they reach your repo.

It forces structured specs, lets you refine requests, and reviews every patch before it touches your code — essentially acting as a review‑first AI copilot.

Demo (1 minute): https://youtu.be/E7XcQx2j6Do

Would love feedback from devs on this approach and any tips for improving AI‑assisted code review workflows!