build-board-packlisted

# Build: /board-pack You are about to generate a `/board-pack` skill customised to this company's leadership reporting. Complete all three phases first. ## Phase 1 — Inspect 1. **Find the governed metric set.** Semantic-layer definitions (dbt metrics, Lightdash `meta`, LookML measures), a metric dictionary if one exists. This is the universe the pack may draw from. If there is no governed set at all, stop and recommend `/build-settle-the-number` first — a board pack built on ungoverned metrics automates the argument, not the answer. 2. **Find the current pack.** Past decks in the repo or linked storage, a template, a recurring doc. Extract its skeleton: sections, metric order, comparison conventions (vs target? vs last month? vs same period last year?), chart styles. 3. **Check data access.** Can Claude Code query the warehouse or the BI tool (MCP, API) to pull the numbers live, or does an analyst paste exports? Both work; the mechanics differ. 4. **Find the calendar.** When does the pack ship, relative to month-end? The gap between data-complete and pack-due defines how much automation is worth. ## Phase 2 — Interview 1. **"Who is the audience, exactly?"** Board, SLT, weekly business review — each reads differently. Board wants five numbers and what you are doing about them; a weekly business review wants the driver tree. The generated skill writes for the named audience. 2. **"Which metrics are the spine of the pack?"** Force a ranking: the five that matter, then the s