Working with multiple Claude Code instances I found issues with requirements traceability, test coverage, and inter-instance communication. I was working with knowledge graphs so built a knowledge graph skill for Claude Code - every user story, use-case, feature, test-case is a node; plus i track defects and enhancements so each instance knows its backlog. It saves tokens by knowing where code is implemented, but perhaps more importantly it found systematic errors in requirements quality and test coverage that I was able to fix. I then extended to support architecture patterns and it will automatically import code and test cases into the graph (using your existing tool). It uses a local Neo4j graph (i run in docker).