aurelienbobenrieth/harness/shopify-review

Shopify review

Scope

• Read surfaces, distribution, deployment, versions, categories from the project. Each surface is its own contract: App Home components don't govern checkout, admin extensions, customer accounts, payments, Liquid.
• Honor project exceptions and deploy restrictions.
• Shopify's official shopify-app-store-review skill and Dev MCP installed → use them alongside this skill, never instead: they check current requirements; this skill owns evidence and Harness tooling. Keep their telemetry off (DO_NOT_TRACK=1, or the toolkit's opt-out file).
• Cite the exact App Store, BFS, App Home, or design section before calling anything an error. Conflicts → record; examples aren't contracts.
• Harness only: pnpm shopify:plan <profile.json> lists pending items from policy/shopify-requirements.json (BFS includes App Store; unreviewed categories stay in); pnpm shopify:sources detects drift, never updates policy.

Evidence owner

• Native HTML, Polaris → scoped lint; reuse upstream a11y rules, no prose regexes or copy autofix.
• App Home on an older Polaris major, or on Polaris React → flag the restyle migration and check its current BFS deadline; 2.0 changes styles, not the component API.
• Manifest, extensions, built output → conformance, explicit deployment. Found script/scope/URL ≠ runtime behavior.
• Copy, UX → scoped agentlint or direct review; findings aren't approval.
• Headers, auth, webhooks, navigation, entitlements, keyboard, layout → app tests. Webhook controls execute app behavior: isolated handlers, fixtures.
• Performance → right app, population, statistic, window; local ≠ dashboard.

Copy

• Merchant task → outcome, consequence, next action; labels, recovery, translations included.
• Voice per situation; mechanics per locale.
• Preserve names, quoted input, consent wording.

Report

Defects, then evidence and gaps. Dashboard, submission, legal, live-store evidence stays unresolved. Green ≠ acceptance or BFS.