When we generate the cluster-manifests, we will use a fixed layout of files. This is good.

However, when we read the cluster-manifests, we currently expect to see the files in the same layout. i.e. the agent-cluster-install.yaml file contains the AgentClusterInstall resource (and only that resource), and so on. This makes it difficult for a user to take a set of ZTP CRs generated from e.g. SiteConfig or some other means and use it to perform an agent-based installation.

Ideally, for the purposes of loading assets we would treat the cluster-manifests as an undifferentiated collection of YAML files. We'd assign resources to specific assets by introspecting their types. We'd validate that there is only one cluster definition, and select between duplicate resources of other types based on the same reference links used in ZTP.

Note that this only affects the installer assets. The files we put in the ISO for the create-cluster-and-infraenv script to read would still need to be in the prescribed format. This will mean changing the way these files are generated (from the individual assets rather than the aggregate Manifests asset).