Are secrets downloaded on a per job basis?

We have a pipeline like this

job A → create AKS cluster (repository_dispatch on repo A)

job B (depends on job A) → deploy apps and run integration tests

job C (depends on job B) → delete cluster from job A

Job A upserts a secret in Repo B that is read from job B

The problem is that Job B fails if the secret didn’t already exists.

Any ideas?

I’d like to avoid KinD or similar clusters if at all possible as part of the tests check Azure infrastructure