databricks

User-specific Databricks rules, the ones that differ from ordinary Databricks practice. Unity Catalog only with no DBFS anywhere, three fixed bundle targets, liquid clustering on every Delta table. Load alongside the vendor databricks-core and databricks-dabs skills, which cover general platform and bundle guidance. This one covers only the deltas, so it does not replace them.

Pause-and-confirm gate between local Databricks development and the real workspace. Use whenever Databricks code (job, pipeline, notebook, app, or anything else shipped to Databricks) has passed local checks - written, locally tested, ruff clean - and the next step would be putting it on the real workspace: validating, deploying to dev, running it, and checking the result there, whatever mechanism the project uses (Asset Bundles, `databricks apps deploy`, workspace import, CLI, MCP, ...). Does not apply if the user's own request already explicitly asked for deploy/run. For how to actually validate, deploy or run, defer to the matching vendor skill (databricks-dabs, databricks-jobs, databricks-apps, ...); this skill only decides whether to ask first.

House style for writing Databricks notebooks and job tasks: DataFrame API only and never Spark SQL, and a fixed cell layout with imports, constants and configs in the first four cells. Use whenever writing or editing a Databricks notebook or job task. For job orchestration itself, task types, triggers and schedules, defer to the vendor databricks-jobs skill. Not for declarative pipelines, see lakeflow-review.

Review a Lakeflow Declarative Pipeline transformation or pipeline resource against Luca's project rules: current dp API spelling, Python pipelines only and never SQL, transformations that stay declarative rather than importable, liquid clustering on every table. Use when asked to review a transformation file, a pipeline definition, or whether a pipeline change follows project standards. For how declarative pipelines work in general, defer to the vendor databricks-pipelines skill. Pair with databricks-conventions.