Currently there is no good way to express it in a formal way:
documenting them as regular attributes could be confusing and lead to collisions when the same attribute is recorded as a regular metric attribute AND as a resource attribute
not documenting them in yaml would be limiting for any automation (md or code generation, validation, etc)
Process and system metrics are expected to be attached to a corresponding resource - https://github.com/open-telemetry/semantic-conventions/issues/1215
Currently there is no good way to express it in a formal way: