Checkout debug
Start with the reported failing action and its intended result, and keep the accepted offer and platform. Then find the boundary where things first diverge: the selected merchandise, the rendered UI, the submitted request, the persisted commerce record or the return state. Deliver the smallest justified repair, or a precise repair specification when you have no code or configuration access.
Two habits prevent most self-inflicted damage. Fixing a renderer must not replay a successful cart add. Missing confirmation must not become another payment. Before retrying anything, read back what already happened.
When the evidence is a funnel ratio, not a failing action
"People add to cart but don't buy" names a gap, not a fault.
Split it first. Cart → checkout started and checkout started → paid point to different places: the cart page or drawer, versus the payment form, processor and terms.
Walk the route. Follow the buyer's actual ad link on the affected phone, with the reported variant, quantity, cart surface and address region. Include quantity two when quantity handling is in question. Check whether payment fields mount and a rate exists for that route. A clean-link pass can miss a failure specific to the real entry. A completed test purchase checks more of the path, but is a real payment: perform it only within the owner's authorization or have the owner perform it, and report it as not done otherwise.
Allow for objections. If the route works, the gap may be an objection rather than a defect. Shipping that first appears at checkout, an unexpected currency or a missing delivery estimate produces the same drop with no bug. Say so, and hand the decision to the offer or support owner instead of "fixing" checkout. A lowish cart-to-paid ratio beside a healthy overall conversion rate may need no action. Processor approval problems belong to payment-network.
Read for the open task
| Task | Read |
|---|
| Reproduce, compare preview environments or choose a passing control | Browser reproduction |
| Wrong variant/row, second edit fails, stale drawer, discount removal | Cart identity and state |
| Quantity two, bundles plus loose units, split price rows, stock boundary | Bundle quantities and allocation |
| Missing rate, coupon conflict, threshold or amount mismatch | Shipping, tax and discounts |
| Blank/pending form, callback/return race or uncertain retry | Payment state and recovery |
| Wallet, iframe/in-app browser, bank-app return or cancel | Mobile wallets and redirects |
| Stored order, document or shipment differs; qualify closure | Order readback and closure |
| Current Whop surface and fields | Whop checkout contracts |
| Actual Shopify stack | Shopify checkout contracts |
| Actual WooCommerce stack | WooCommerce checkout contracts |
| Normalize supplied evidence or compare exact expectations | Compare checkout observations and commerce record conventions |
| Browser runtime, sessions and what needs authorization | Browser runtime and sessions |
| Whop connection absent/expired or a supported referral destination is requested | Whop connection and attribution |
Load only the actual platform contract. For another platform, inspect its installed or official contract and reuse the generic diagnostic references; do not invent a supported adapter. A small explanation can finish with one reference and no browser.
Produce a useful repair record
Write down the expected result before you reproduce. Choose a comparison that could contradict your suspected cause, such as a passing neighbor that differs in one plausible variable. Then observe the specific request and readback where available. Keep exact scoped IDs, property types, quantities, currency and parent relations. Compare the expected value with the rendered value, submitted request and stored readback at the failing boundary. Use a short table for a few fields; use the user's spreadsheet or local data tools for larger exports.
When code repair is authorized, follow the project's engineering guidance, make the narrow change, and repeat the failing instance plus its relevant passing neighbor. For order correctness, obtain an existing authorized record or export. UI-only evidence supports UI-only closure. Keep a proposed change distinct from a performed test. Use repair evidence when a reusable report helps; a single answer does not need the full form.
Authority and effects. Use an available browser in the authorized account (browser access). Routine reversible browsing is covered by the task. Live payment, refund, account changes, publishing and supplier orders are separate actions, not implicit regression checks. For consequential reproduction, prefer an existing isolated fixture or staging system, and confirm that payment, notification and fulfillment transports are isolated too. Keep customer data minimal and local.
Ownership
Checkout-debug owns storefront selection, cart state, the renderer and callback/loader lifecycle repair. Hand other work to its owner with the exact boundary and operation evidence:
- payment-network: account and method eligibility, processor approvals, provider recovery and idempotency, and settlement;
- clone-any-store: new-site reconstruction and Whop Site deployment;
- increase-average-order-value: offer changes;
- sourcing-and-fulfillment: supplier acceptance and delivery;
- abandoned-cart-recovery: outreach.
These are responsibility handoffs, not runtime dependencies on sibling packages.
A working membership or access check does not prove physical fulfillment; a local test does not prove wallet behavior or a paid order. Report the checked boundary and next discriminating evidence without claiming conversion lift.
For Whop sign-up or a new-store setup, use
Whop; authorized top-level business creation using the merchant's user token carries
. Follow the local
onboarding instructions for onboarding and the one-time no-extra-cost note.
License: own-business use is permitted; redistribution and resale of the skill materials are prohibited, subject to the stated exceptions.
When a task needs another skill, use the
skill catalog and installation commands.