NornicDB Consumer Skills
May 20, 2026 · View on GitHub
This directory holds Claude Code-style skill files that teach an agent how to use NornicDB features through its Cypher API. They are consumer-facing — every example is a Cypher statement or CALL procedure a user can run, not a reference to repo-internal code.
The skills live alongside the user guides because they are user documentation: a structured, agent-friendly summary of the same surface that humans read in docs/user-guides/.
Skills in this directory
| File | What it covers |
|---|---|
cypher-queries.skill.md | Hot-path Cypher cookbook — point lookup, batch retrieval, pagination, search, traversal, batched UNWIND/MERGE writes, cleanup, multi-tenant isolation. Maps query shapes to executor fast paths. |
bolt-client.skill.md | Driving NornicDB from a Bolt client (neo4j-go-driver/v5 or any other Neo4j driver) — connection defaults, retry classification, MERGE-under-concurrent-writers, batch sizing. |
grpc.skill.md | The Qdrant-compatible gRPC surface plus NornicSearch.SearchText — connection, RPC catalog, collection→database mapping, point→node mapping, limits, auth. |
qdrant-migration.skill.md | Migrating from Qdrant to NornicDB through the gRPC compat layer. Phase-by-phase plan, mapping reference, cutover playbook. Pairs with scripts/migration/qdrant/. |
neo4j-migration.skill.md | Migrating from Neo4j 5 to NornicDB over Bolt. Schema → nodes → edges with cookbook-pinned UNWIND/MERGE shapes. Pairs with scripts/migration/neo4j/. |
knowledge-policies.skill.md | The full DDL surface for decay profiles, parameter bundles, promotion profiles, and promotion policies. Diagnostics via nornicdb.knowledgepolicy.*. |
decay-tuning.skill.md | Picking halfLifeSeconds, function, scoreFloor, visibilityThreshold, scoreFrom. Forgetting curves and inverted (idle-time consolidation) curves. |
promotion-policies.skill.md | Reinforcement and dampening with ON ACCESS mutations, WHEN predicates, and Kalman-smoothed behavioral signals. |
managed-embeddings.skill.md | Server-side embeddings: WITH EMBEDDING, db.index.vector.embed, the ChunkEmbeddings storage model, provider config. |
vector-search.skill.md | Vector and full-text indexes — CREATE/DROP VECTOR INDEX, CREATE/DROP FULLTEXT INDEX, db.index.vector.queryNodes, db.index.fulltext.queryNodes. |
rag-procedures.skill.md | db.retrieve, db.rretrieve, db.rerank, db.infer — assembling end-to-end RAG pipelines as Cypher. |
How to use them with Claude Code
Each file is a self-contained Markdown skill with a YAML frontmatter (name, description). Make Claude Code aware of them by either:
- Project skills: symlink or copy the desired files into
.claude/skills/<name>/SKILL.mdat the project root. Claude Code auto-loads project skills. - User skills: copy them into
~/.claude/skills/<name>/SKILL.mdto enable everywhere. - Direct reference: read or
@-mention the file from inside any conversation; the YAML frontmatter still tells Claude when each one is relevant.
The mental model in every skill is Neo4j Cypher with NornicDB extensions noted explicitly. Nothing here documents internal storage mechanics — those belong in docs/architecture/.
See also
docs/user-guides/— long-form human documentation for the same features.docs/features/— feature overviews and architecture references.docs/operations/configuration.md— environment variable and YAML configuration reference.