ship-verify-with-prod-datalisted

Before shipping a change to a data-facing surface — a dashboard, a report, a total, a count, a funnel, a chart, any figure whose job is to reflect stored data correctly — prove it renders correctly and will NOT regress against REAL production data, across EVERY prod account with recent data, not one. Reads prod READ-ONLY through the provisioned path, enumerates every account/tenant/entity the surface serves, recomputes what each figure would show under the new code, and confirms they cross-verify to one source. Distinguishes a real code bug from a deploy/data-skew (prod written by the old code) or legacy residue (stale records the current writer no longer produces, dated to prove it), supplies a backfill or recompute the user runs on prod, and proves that backfill first on a seedable dev account. Use for "verify against prod data", "check it works for all accounts with real data", "make sure this won't regress on production", "look at recent data from all accounts", "does this add up across accounts", or befo
skowalskidev/dotclaude · ★ 2 · Data & Documents · score 78

Install: claude install-skill skowalskidev/dotclaude

# Verify a data surface against real production data — for ALL accounts, before shipping A green test suite proves the code runs on fixtures. It does not prove the numbers are right on the real data every account actually has. A money/data surface can pass every test and still read `$0`, double-count, or disagree with itself on a live account whose data shape the fixtures never had. This skill closes that gap: read prod read-only, recompute what the surface WOULD show for every account with recent data, and confirm it reconciles before the change ships. It composes, and does not restate: - **`~/.claude/rules/connectors.md`** — the read-only prod path, the auth-gate protocol, the work/personal boundary. Prod is READ-ONLY; a prod WRITE needs Simon's per-write go-ahead. - **`/sk:ship-screenshot-changes`** — the capture + PR post (Phase 6 hands off to it). - **`/sk:ship-report-and-ensure-correct-user-system-journey`** — the acceptance-criteria verdicts this feeds a real-data verdict into. ## Phase 1 · Decide whether it applies **DO run this when the diff changes how a surface READS, derives, reconciles, or enumerates stored data** — a displayed total/count/rate, a funnel, a chart, a cross-surface reconciliation, or the list of entities a sweep/report covers. **DON'T run it for a pure-logic change with no data-shaped output** (a refactor, a copy tweak, a new pure function with unit tests). Say it doesn't apply and stop. TEST: name the exact figure(s) on the surface whose c