How do you prevent a change in a helm template to be deployed to one cluster (say, production) but make sure it will be deployed to another (say, stage)?

See the multi-cluster example “helm” on here: https://github.com/rancher/fleet-examples/

So basically using conditional values within the templates?

Yes, in

fleet.yaml

you can see the values for dev, test, prod: https://github.com/rancher/fleet-examples/blob/master/multi-cluster/helm/fleet.yaml

Yes that's how I do it already, but it feels so sketchy. One mistake and everything breaks

I get that. If you feel it’s too risky, you can instead just duplicate your code and deploy separately to the envs to be on the safe side. Or have another test environment for fleet deployments itself.

Thanks @nutritious-orange-38459