build-training-programmelisted

# Build: your analyst training programme + /coach You are about to generate a training programme customised to this team — session plans built on their actual backlog, plus a `/coach` skill for practice between sessions. Complete all three phases first. ## Phase 1 — Inspect (find the real practice material) 1. **Find the safe sandbox.** Which parts of the repo can a trainee touch without risk? Look for: dev/staging targets in `profiles.yml` or CI config, non-production dbt targets, draft spaces in the BI config. The generated exercises happen there. 2. **Harvest real exercises.** Search for the backlog in code form: `TODO`/`FIXME` comments, models missing tests, `schema.yml` files with empty descriptions, charts referencing renamed columns. Every one is a training exercise with real stakes and a mergeable outcome. List 10–15 candidates, easiest first. 3. **Check what skills already exist.** `.claude/skills/` in their repos — if `/fix-chart` or `/new-model` are already installed (perhaps from this foundry), the curriculum trains analysts on *their* skills, not generic prompting. 4. **Gauge the stack's teachability.** dbt + semantic layer present? Then the curriculum can promise the 90-minute first ship. No semantic layer? The curriculum starts a step earlier, and the generated programme says so honestly rather than overpromising. ## Phase 2 — Interview 1. **"How many analysts, and what is their SQL comfort, honestly?"** (Can review a diff / can write a CTE / SELECT-star-a