Spec

Prerequisites

Dashboard linking v1 (initiative Dashboard linking) should ship or be far enough along that cross-board URLs are stable; program setup depends on it. Research: Dashboard linking — research.

In: Open-source Fivetran quickstart dbt projects (one git repo per connector pack), Dataface YAML dashboards, navigation patterns, research artifacts, agent skill for repeatable research.
Out: Changes to upstream dbt packages (unless filed separately); proprietary vendor asset reproduction; full “product clone” UI parity.

Choose a root directory (example: sibling to this repo, ../quickstart-dbt-workspace/, or $QUICKSTART_DBT_WORKSPACE). Do not assume paths inside dataface/ without documenting ignore/submodule policy.
For each connector: git clone <quickstart-repo-url> <connector-slug> (exact naming convention TBD in program task—prefer connector slug from Fivetran naming).
In each checkout: run dft init from the dbt project root (or documented subdirectory if the repo is monorepo-style).
Dashboard YAML lives in that project following whatever dft init establishes (document any overrides per repo).

Clone + dft init.
Run quickstart-product-dashboard-research skill: capture native dashboards, question list, entity/detail topology, link matrix, model gaps.
Implement a minimal first pack: executive/overview board + one entity-detail path with cross-links.
Record retro notes in the pilot task; open new tasks for follow-on connectors or skill revisions.

Every chart maps to a real modeled relation; no silent aggregation magic.
Boards that claim “vendor parity” cite research notes (section + link), not memory.
Cross-board links use stable keys present in the quickstart model (document key choice in link matrix).

dft serve (or project-standard serve) renders the pack against a documented seed / sample database (pilot task defines which).