You've got Claude Code running in terminal. Cursor open in one window. Maybe Windsurf too. Each one doing exactly what you told it — in three separate rules files that haven't been in sync since February.
This is the AI coding governance problem. It's not talked about much because most people don't notice it until an AI agent refactors a function the wrong way, or commits secrets, or breaks your naming convention for the fourth time this month.
The rules are there. They're just scattered.
What's Actually Happening
Every major AI coding tool has its own rules format:
- Claude Code reads
CLAUDE.md - Cursor reads
.cursorrules - Windsurf reads
.windsurfrules - Copilot has its own instructions format
Most teams maintain one of these seriously and let the others drift. So Claude Code knows your stack preferences. Cursor doesn't. You update .cursorrules for a new project convention. Forget to update CLAUDE.md. Three weeks later you're debugging why your two AI tools are giving contradictory suggestions on the same codebase.
It's not an AI problem. It's a config management problem.
What Crag Does
Crag is a governance layer for AI coding tools. You write one source-of-truth rules file, and Crag syncs it across every tool format in your repo.
Change one line in your central config — Crag propagates it. Switch from Cursor to Windsurf on a project — your rules come with you. Onboard a new dev — they get the full governance context regardless of which tool they prefer.
The 50-repo test that's been floating around showed 96.4% translation accuracy. That's not perfect, but for most practical rules — naming conventions, stack preferences, security constraints, commit message formats — it holds up well. The failure cases tend to be tool-specific features that don't have equivalents across platforms, which Crag flags rather than silently mistranslating.
That's the right behavior. Silent mistranslation is how you end up with an AI that thinks it understood your rules when it didn't.
Who Actually Needs This
If you're running one AI coding tool on one project, Crag is overkill. You have one rules file. Maintain it.
Crag earns its place when you're:
Running multiple tools in parallel. Some people use Claude Code for agentic tasks (long multi-file changes, refactors) and Cursor for quick edits. If that's you, you're already managing two rule sets — Crag collapses that.
Working across multiple repos. If your rules are mostly the same across projects but you're copy-pasting and manually tweaking, Crag's inheritance model handles this cleanly. Central base rules, per-project overrides.
Onboarding AI to a team codebase. Getting a team's rules consistent is harder than getting your own consistent. A single authoritative file that generates all the tool-specific outputs removes a whole category of "why did the AI do that" debugging.
Frequently switching tools. The AI coding space is moving fast. Cursor was the obvious choice two years ago. Now Claude Code is genuinely competitive for certain workflows. If you're evaluating tools regularly, portability has real value.
Getting Started
The setup is straightforward:
- Install Crag in your project
- Create a single governance file (Crag has a schema — it's not freeform, which is a good thing)
- Run the sync command — it generates your
.cursorrules,CLAUDE.md,.windsurfrules, and whatever else you need - Add the sync to your workflow (pre-commit hook, CI step, or just run it manually when you update rules)
The schema-based approach is worth noting. You're not writing free-text rules and hoping Crag interprets them correctly — you're writing structured config that Crag knows how to translate precisely. That's what gets you the high accuracy numbers.
Honest Take
The governance problem is real. Crag's solution is the right shape of solution.
The part I'd watch: the tool is only as useful as your rules file is good. If your existing .cursorrules is three vague lines about "write clean code," syncing it everywhere doesn't help you. The forcing function of writing a proper Crag config is actually the most valuable part for a lot of teams.
Also worth knowing: this category is going to get crowded. The big AI coding tools will eventually standardize on something — or someone will build the equivalent into an IDE. Crag is solving a real problem today. Whether that window is one year or three, I can't tell you.
For now, if you're managing multiple AI tools across multiple repos and your rules are drifting, it's worth ten minutes to evaluate.
Written by McKlaud AI. Want to know which AI tools actually fit your business? Get a free AI audit.