Ontology Knowledge Graph Management

Problem

Knowledge graphs often break down between data entry and retrieval: after adding entities, relations, or documents, teams can encounter glued JSON lines, broken quotes or brackets, orphan nodes, and inconsistent SQLite synchronization. This skill addresses those engineering details by turning graph completion, validation, synchronization, and vector search into an executable workflow rather than a conceptual outline.

How it works

The main capabilities cover several lifecycle tasks:
- Reading and validation: it reads raw graph data and checks common format issues, such as Line N: double JSON and invalid quotes or brackets, then suggests splitting or repairing them.
- Completion management: it supports entity, relation, and document formats for appending structured data to an existing ontology graph.
- Sync and verification: it writes data to SQLite and checks entity counts, relation counts, and optionally orphan nodes.
- Vector search integration: by default it can use SQLite plus in-memory vectors, and can connect backends such as ChromaDB, Pinecone, or Weaviate depending on deployment needs.
- Index refresh: it includes scheduled index updates to reduce stale retrieval results.

Boundaries and cautions

The skill references external scripts such as scripts/graph_sync.py, scripts/graph_vectorize.py, and scripts/graph_search.py, but they are not included in the current directory. Obtain them from the project repository or implement them against the described interfaces. Before running, confirm no partial operation is in progress; back up configuration before changes; and verify end-to-end behavior after completion. For vector storage, ChromaDB is suitable for local small-scale use, while Pinecone or Weaviate can fit cloud or larger deployments.