Conciseness

The body is information-dense and mostly earns its tokens, but it repeats itself: session-label reuse guidance appears three times (prose paragraph, code comment, and the 'Picking a session label' bullet), and sections like 'Relationship to the Genie One MCP' and the Genie/OneChat naming history add context Claude rarely needs to act. This matches 'mostly efficient but could be tightened' rather than the lean 4–5 anchors.

Actionability

Commands are copy-paste ready and cover the common cases: full `databricks genie ask -s trips "..."` examples for follow-ups and summarization, `--include-sql`, `--output json` with the actual JSON shape shown, the legacy `experimental genie ask` fallback spelled out, and the exact verification command (`databricks genie ask --help`) before falling back. This matches the 'fully executable, copy-paste ready, specific examples cover common cases' anchor.

Workflow Clarity

The routing decision (Genie vs coding agent) is laid out first with explicit do/don't categories, the fallback branch has a real validation checkpoint (verify the command exists before assuming it's missing), and error recovery is handled ('a non-answer is a message, not an error: don't retry — use the manual fallback'; stalled-stream failure behavior). It sits between the 4 and 5 anchors: checkpoints exist but are scattered across prose rather than a sequenced checklist, and the anti-pattern guidance ('do NOT default to tables list') is stated without a recovery step.

Progressive Disclosure

Well-structured single-file body with clear section headers routing from overview to usage details to fallback, and one clearly signaled one-level-deep reference to the parent skill's manual-data-exploration.md. It falls short of the 5 anchor because the detailed flag/output documentation (parallelism rules, JSON schema, session-label mechanics) is inlined and would fit naturally in a reference file, and no reference/script files exist in the bundle to carry that depth.

Specificity

"Discover, explore, and query Databricks data via Genie" names three concrete capabilities plus the mechanism (CLI, Genie One), matching the 'several specific actions; minor gaps' anchor. Not a 5 because 'discover/explore' remain somewhat abstract compared to fully concrete action verbs, and the Databricks-specific operations (session-labeled asks, SQL generation) are only implied by the trigger examples.

Completeness

Both questions are answered explicitly: what — "Discover, explore, and query Databricks data via Genie — the CLI equivalent of the Genie One MCP"; when — "MUST be invoked whenever the user asks to find or locate data (...), answer a natural-language question about the data, or write a SQL query", with concrete trigger phrases. This matches the top anchor exactly; nothing is merely implied.

Trigger Term Quality

Strong natural-language triggers are quoted directly — "what tables are in X", "where does X live", "which catalog/schema has Y", "answer a natural-language question about the data", "write a SQL query" — covering how users actually phrase these requests. A few natural variations (e.g., 'find data', 'profile a table', 'schema') are absent, so it falls just short of the comprehensive-synonyms anchor.

Distinctiveness Conflict Risk

The Databricks/Genie niche is clear and unlikely to collide with unrelated skills, but the unqualified trigger "or write a SQL query" is broad — a user asking for SQL against a non-Databricks database would match the trigger text literally, creating minor overlap risk with generic SQL-authoring skills. That keeps it just below the 'minimal conflict risk' anchor.