← ClaudeAtlas

governed-report-contractlisted

Defines and validates a versioned governed report contract - report.yaml plus parametric trusted queries and metric-view bindings - so a report returns the same numbers in a skill, a script and a Databricks App. Use when the user wants trusted, pinned or reproducible report queries, or to validate one. Not for creating metric views, Genie agents, AI/BI dashboards or app scaffolding.
Paldom/databricks-skill-to-app · ★ 0 · Data & Documents · score 71
Install: claude install-skill Paldom/databricks-skill-to-app
# Governed report contract A report becomes ungovernable the moment its SQL lives inside whatever is rendering it. The skill gets one version of the numbers, the app gets another, and nobody can say which is right. This skill makes the SQL the artifact: one versioned directory of parametric, read-only queries over a Unity Catalog semantic layer, validated by a script, that every consumer reads instead of inventing its own. **What the contract does and does not promise.** It guarantees that every consumer runs the *same reviewed query definition* with the *same declared parameters and execution identity*. It does not freeze the data or the semantic layer: the metric view can be redefined and the source can load between two blocks. That is why the contract carries a freshness watermark and per-block execution metadata instead of claiming a report-wide as-of time. `trust: certified` means "deterministic SQL the owner reviewed" — it is a review status, not an attestation. The contract is a convention defined here, not a Databricks product feature. It deliberately reuses the platform's own file formats — AppKit `-- @param` annotations, the `.obo.sql` naming rule, AppKit's `metric-views/definitions.json` — so materializing it into an app is a **copy, not a translation**. ## When to use - The user wants trusted / certified / pinned queries behind a report. - The same numbers must appear in more than one place (a skill, a scheduled script, an app). - A report's numbers moved and