dex

Use this to make sense of a database, warehouse, or DuckDB file: inventory and rank what is there, profile columns, detect PII, flag grain and data-quality problems, infer and verify how tables join, answer ad-hoc data questions with guarded SQL probes, and cluster rows into segments with k-means, producing a draft map without dumping the schema into context. Trigger it for casual, artifact-first prompts like "what's in my duckdb", "what's in this database", "what data do I have", "take a look at data.duckdb", or "any PII in here", as well as analyst questions like "what is in this warehouse", "which tables matter", "what does this table contain", "how do these tables relate", "is this data any good", "profile these columns", or ad-hoc counts and distributions like "how many orders have no customer", and unsupervised segmentation like "cluster my customers", "find natural segments in this table", or "run k-means on these columns". Any mention of exploring, inspecting, querying, understanding, or clustering a

Use this to keep a dbt project correct as the warehouse and the business change. It detects drift on four axes and proposes the fix: schema drift (source columns and tables added, dropped, retyped, or renamed), volume drift (a row count that collapsed, a table that emptied, a load that half-failed), grain drift (a key that lost uniqueness, a changed row-per-entity cardinality, an increased join fanout), and semantic drift (a metric, measure, dimension, or entity definition that no longer matches, new categorical values, dangling semantic references). Trigger it for requests like "what changed in the warehouse", "did anything drift", "is my dbt project still in sync", "my primary key has duplicates now", "the row count dropped", "did the load run", "the data stopped flowing", "the revenue metric definition changed", "reconcile my models with the source schema", or "which models are stale". It reads the .dex/ snapshot and proposes reviewable diffs; it never overwrites hand-written work. Do not use it to author

Use this to author and change a dbt project: bootstrap a new dbt project in a repo that has none (`transform init`), write or refactor dbt model SQL from staging to marts, add tests and docs in schema.yml, manage dependencies, and define or update the semantic layer (dbt semantic models / MetricFlow: entities, dimensions, measures, metrics). Trigger it for requests like "set up a dbt project in this repo", "build a staging model for this table", "refactor this model", "add tests to this model", "create a mart for X", "define a revenue metric", or "add a dimension to this entity". Every change is a reviewable diff to the dbt project; any warehouse build is dev-target only, gated, and cost-surfaced first. Do not use it to explore or profile a warehouse (use explore) or to detect drift and reconcile a project that has fallen out of sync (use maintain).