Changelog
2026-07-31
Added — 3 AOV report skills (856 → 859)
(The jump from 844 to 856 between the two entries below isn’t a bookkeeping error: the Mon/Wed/Fri cron on the Mac mini shipped 12 skills to main in four commits while this session was running, and they arrived here on a rebase.)
Batch sources/new-aov-reports-2026-07-31.json, ingested via bun _tools/skills.ts ingest.
sidekick-aov-by-payment-method-report— AOV split by payment method. Asks Sidekick which breakdowns the store’s analytics actually supports before running; the BNPL row is conditional on it being isolable.sidekick-aov-paid-vs-free-shipping-report— AOV of orders that paid shipping vs orders that shipped free.sidekick-gift-card-aov-impact-report— AOV of gift-card-redeeming orders vs the rest, gated on Sidekick first confirming redemption links to orders at all.
All three: Analytics / Medium / executable. Verified with bun _tools/skills.ts lint (859 skills, no structural errors) and a Liquid grep over the three new files.
The shipped set differs from what the researcher proposed. The auditor rejected sidekick-aov-by-shipping-method-report outright — ShopifyQL has no shipping-method dimension; sales exposes shipping_charges as a metric only — and wrote a different job in its place rather than patching a prompt built on a dimension that doesn’t exist. It also stripped an “average top-up” table from the gift-card report for the same reason: no schema exposes a per-order applied amount.
Known gap: sources/_capability-rubric.md §2:67 is the only ShopifyQL line in the rubric and it enumerates zero dimensions. Auditing report prompts therefore means going to shopify.dev for the sales / payments / gift_cards schemas every time. The rubric should carry that dimension list.
Added — 3 Customer Loyalty skills (841 → 844)
Batch sources/new-loyalty-2026-07-31.json, ingested via bun _tools/skills.ts ingest.
sidekick-loyalty-perks-catalog-metaobject— create a “Loyalty Perk” metaobject type and one entry per tier/perk. A static reference catalog, explicitly not a points balance.sidekick-segment-tier-upgrade-nudge— segment customers within a spend gap of the next tier, tag them, and draft (never send) a Shopify Messaging email naming the benefit they’d unlock.sidekick-reconcile-tier-tags-vs-spend— reconcile existing tier tags against actual spend.
All three: Customer Loyalty / Medium / executable.
Process: skill-researcher (Sonnet) → skill-auditor gate (Opus) → ingest. The researcher delivered 2 of 3 and refused to pad — Customer Loyalty already held 55 skills, and the rubric bars points ledgers and third-party loyalty apps. The auditor verified that reasoning, then wrote the third itself against a confirmed native capability.
Verified with bun _tools/skills.ts lint (844 skills, no structural errors) and a Liquid grep over the three new files. bundle exec jekyll build could not run locally — the github-pages gemset isn’t installed on this machine (pre-existing Bundler::GemNotFound). The GitHub Actions build on push succeeded instead: all three pages return 200 live.
Known gap: .claude/agents/skill-researcher.md lists gift cards as a strong zone, but sources/_capability-rubric.md §2 never confirms gift-card creation or issuance. One candidate prompt was rejected over it. The rubric should settle this either way.