I’ve been using AI coding tools like Cursor and Claude quite a bit, and one recurring issue I ran into was this:
Once code is generated, there’s very little control over what actually gets changed in the codebase.
Sometimes the AI: • modifies unrelated files • introduces changes outside the intended scope • or drifts from the original structure
Most tools either help generate code or review it after the fact, but there’s no real control layer in between.
So I built a CLI that sits between AI-generated code and your repo, and tries to: • define boundaries (which files should or shouldn’t change) • check diffs against those constraints • flag or block changes that violate them
The idea is to enforce constraints deterministically, instead of relying on another AI pass to review generated code.
Recently open-sourced it — still early, but would love feedback from folks actively using AI tools in their workflow. We have 20K+ CLI installs and 1K+ IDE extension installs as of now.
Curious how others are dealing with this today.
Comments URL: https://news.ycombinator.com/item?id=47696831
Points: 3
# Comments: 1