For several months, I noticed the same problem with Cursor, Windsurf, and other AI assistants. They're excellent…
but they quickly forget project rules.
At first:
clean architecture
conventional commits
documentation
tests
Then, after a few hours of development:
200-line functions
direct push to main
no ADR
forgotten documentation
ignored OWASP rules
End up, you end up repeating the same instructions:
"Use GitHub Flow"
"Don't forget the tests"
"Respect the complexity limits"
"Document this decision"
Again.
And again.
And again.
So I decided to reverse the problem.
Instead of reminding the AI of the rules…
...why not integrate them directly into the project?
This is exactly what Guardian Engineering Framework (GEF) does.
The idea
GEF is a project generator.
But unlike a simple scaffold, it also installs:
Engineering Playbook
AI prompts
Git Hooks
CI/CD
Cursor/Windsurf rules
ADR
Diátaxis documentation
right from project creation.
npx create-gef
A few minutes later, the project already has technical governance.
What is generated
GEF automatically configures:
GitHub Flow
Conventional Commits
Pull Request template
CI GitHub Actions
Docker
documentation
ADR
lint
security
AI prompts
The goal is for the project to enforce best practices, even when multiple developers or multiple AIs are involved.
An AI that acts as a Tech Lead
This is the point I find most interesting.
The framework pushes the AI to:
create ADRs
open Pull Requests
respect issues
run tests
follow a reproducible workflow
The AI is no longer just a code generator.
It follows a process.
Why it's open source
I wanted something usable with:
Cursor
Windsurf
GitHub Copilot
Claude Code
without depending on a specific IDE.
All the rules reside within the project.
Repository
GitHub:
https://github.com/Gnzikoune/GEF NPM:
https://www.npmjs.com/package/create-gef Feedback is welcome.