build-dbt-modellisted

# Build: /new-model You are about to generate a `/new-model` skill customised to this company's dbt project. The skill must produce models that look like they were written by the team's most careful engineer — which means this builder's real job is working out what "most careful" means *here*. Complete all three phases first. ## Phase 1 — Inspect (the conventions live in the code, not the docs) Sample generously — at least 15 models across layers. Extract and record: 1. **Layering.** `staging → intermediate → marts`? Medallion? Something home-grown? Check `dbt_project.yml` folder config and actual directory structure. Note the naming per layer (`stg_`, `int_`, `fct_`, `dim_`, `mrt_`?) and whether it is applied consistently or aspirationally. 2. **SQL style.** CTE structure (import CTEs first? one final `select`?), leading vs trailing commas, capitalisation, `ref()`/`source()` discipline, jinja usage, macros the team actually uses. Copy two representative models — the generated skill will embed them as golden examples. 3. **Testing reality.** What share of models have `unique`/`not_null` on primary keys? Which custom/dbt-utils tests appear? The generated skill enforces the standard of their *best* models, not their average — but name the gap honestly in your summary. 4. **Documentation reality.** `schema.yml` coverage, description quality, whether descriptions predate the SQL (`git log` both — a description older than the SQL it describes has already gone stale). 5. **Mate