Three months of using AI coding assistants will teach you one thing: the AI will modify files you didn't ask it to touch. One developer got tired of it and built a fix.
The project is a command-line tool that sits between AI coding assistants like Cursor and Claude Code and your actual repository. You define which files and directories are off-limits, and changes outside those boundaries get blocked before they reach your codebase. Think of it as a .gitignore for AI edits - except instead of hiding files from version control, you're restricting which parts of the repo an AI agent is allowed to modify.
The problem is real. AI coding tools are optimized for generating useful code, not for surgical precision about what they change. Ask Claude Code to refactor a utility function and it might also "helpfully" update your config files or adjust imports across unrelated modules. For solo projects, this is annoying. For team codebases with files that shouldn't be touched without a proper review process, it's a genuine problem.
Current tools handle this with post-hoc review - you check the diff after the AI runs. This CLI moves control upstream, stopping unwanted changes before they happen rather than catching them after.
It's a one-developer project at this stage, so integration with specific AI coding tools may be rough around the edges. But the underlying need - more granular control over what AI agents can actually touch in your repo - is something the major tools haven't fully addressed. If the pattern works, expect to see it absorbed into the main tools eventually.