Conciseness

The body is lean and assumes competence — no space is spent explaining what Kafka or Neo4j is, and configs are shown as compact JSON blocks. Not 5: the full 'Event Structure' block and 'Selector Reference' table duplicate material that references/cdc-api.md explicitly covers ('event structure detail'), so some tokens don't earn their place; not 3: outside that duplication there is no unnecessary explanation.

Actionability

Copy-paste-ready content throughout: complete sink/source connector JSON configs, executable Cypher (EOS constraint, db.cdc.query variants), a working Python cursor-loop, and a Common Errors table mapping error → cause → fix. Not 4: the examples cover the common cases end-to-end, including the Confluent Cloud variant and schema-registry converter settings.

Workflow Clarity

A decision table routes users to the right strategy, prerequisites state version constraints, EOS is a numbered 2-step sequence (constraint then config), and the closing checklist plus error-recovery table provide validation checkpoints for batch/database operations. Not 5: validation is concentrated in a terminal checklist rather than inline checkpoints (e.g., no explicit 'verify connector task is RUNNING / consume one message' step after deploying a config); not 3: the cap does not apply because validation steps are explicitly present.

Progressive Disclosure

Good structure with clearly signaled, one-level-deep references that exist on disk ('[Full connector config reference](references/sink-config.md) — all neo4j.* properties, defaults, types'; '[CDC API patterns](references/cdc-api.md)'), and the bulk config reference is correctly split out. Not 5: the event-structure and selector-reference material inlined in the body belongs in cdc-api.md, which already advertises that content; not 3: the split and navigation are otherwise well done.

Specificity

Enumerates multiple concrete capabilities — 'Cypher/Pattern/CUD sink strategies, CDC-based and query-based source, exactly-once semantics, DLQ error handling, Confluent Cloud managed connector, schema registry (Avro/JSON), and native db.cdc.query cursor-loop patterns' — with comprehensive coverage of the skill's scope. Not 4: there are no meaningful gaps; every major capability of the connector and CDC API is named explicitly.

Completeness

Explicitly answers both questions: 'what' in the first two capability sentences and 'when' via 'Use when streaming Kafka events into Neo4j, streaming Neo4j changes to Kafka, or querying Neo4j change events without Kafka'. Not 4: the 'when' clause is fully explicit with concrete trigger phrases rather than merely present.

Trigger Term Quality

Natural user phrasings are covered: 'streaming Kafka events into Neo4j', 'streaming Neo4j changes to Kafka', 'querying Neo4j change events without Kafka', plus domain synonyms (Kafka, Neo4j, CDC, Avro, schema registry, DLQ). Not 4: the three 'Use when' clauses capture the distinct ways a user would phrase each intent, and no file extensions are relevant to this domain.

Distinctiveness Conflict Risk

Clear niche (Neo4j-Kafka streaming and CDC) with explicit negative boundaries redirecting to sibling skills: 'Does NOT handle Cypher query authoring — use neo4j-cypher-skill', 'Does NOT handle bulk CSV/file import — use neo4j-import-skill', 'Does NOT handle GDS algorithms — use neo4j-gds-skill'. Not 4: the boundary clauses make overlap with related skills nearly impossible.