Knowledge graphs (KGs) continue spreading into industrial use cases due to their advantages and superiority over classical data representations. A problem that has not yet adequately been solved for KGs is the traceability and provenance of changes, which can be required in an enterprise or by regulations. KGs typically contain the current snapshot of data valid at a certain moment in time only. Changes over time are usually not recorded and no change history exists. The lack of suitable and scalable traceability solutions hinders the wider application of KGs. This paper presents a traceability and provenance solution for KGs, which can track all changes of a KG on triple level. It comprises a provenance engine that intercepts SPARQL/Update queries; PROV-STAR, an RDF-star enabled light-weight extension of the Provenance Ontology (PROV-O) for representing changes and their provenance; and a SPARQL query transformation approach for tracking the changes on a separate provenance KG with SPARQL-star queries. The solution supports full traceability of all changes, on the lowest possible level of triples, with each change being comprehended with detailed provenance information. From the provenance KG a detailed change history can be retrieved, and any past version of the KG can be restored with a single query. The implementation and validation have shown that changes can be tracked at runtime during the normal operation of a KG. Furthermore, the solution is scalable to large KGs and frequent updates, as only the delta of each change is stored.