Test admin draft order creation and mark as paid

The flow under test

Admin flows are stateful in a way storefront flows are not. The test has to establish what was there before, change it, and then prove the change landed in the right place with the right attribution.

• Establish the baseline - Orders (2), listing #1001 and #1002, so the new order is demonstrably new.
• Build with real inputs - a named customer, an email, a selected product and a quantity of 2 producing a $52.00 draft total.
• Check the intermediate state - draft D1 exists for Marco Oduya with status open before anything is charged.
• Check the attribution - order #1003 shows channel "Draft order" and status paid, not just that an order exists.

The channel assertion is the detail that makes this test worth running. An order that appears but is attributed to the wrong channel corrupts every downstream sales report, and it looks completely fine on screen.

The test file

Step three does the real work: five separate interactions, including a dropdown selection, in one instruction. Writing it as one step keeps the file readable and lets Kane CLI plan the form in a single pass.

---
mode: testing
url: https://my-testing-repo-main.vercel.app/shopify-clone-app/admin?reset=true
max_steps: 45
tags: [shopify, admin, orders]
---

# Shoplify 2.4: Admin order creation

Catalog objective: create a draft order for a customer in admin and mark it paid.
Key assertion: the order appears in the orders list as paid.

## Confirm the starting orders list
Verify the Orders tab shows "Orders (2)" listing order "#1001" and order "#1002".

## Open the draft order builder
Click "Draft orders" and verify a "Create draft order" card is shown with a "Customer name" field.

## Build the draft
Type "Marco Oduya" into "Customer name", "marco@example.com" into "Customer email", select "Canvas Market Tote" as the Product, type "2" into Quantity, click "Add line item", and verify a draft total of "$52.00" is shown.

## Save the draft
Click "Save draft order" and verify the draft orders list shows draft "D1" for "Marco Oduya" with status "open".

## Mark the draft paid
Click "Mark as paid" on draft "D1" and verify the Orders tab shows order "#1003" for "Marco Oduya" with channel "Draft order" and status "paid".

Run it with:

kane-cli testmd run tests/shopify/admin-order-creation_test.md

The run

All five passed. The build step took 141.8 seconds, half the total run, which is the cost of a multi-control form with a product dropdown and a line item that has to be added before the total appears.

---
test: ../admin-order-creation_test.md
status: passed
started: 2026-09-15T09:29:05.519Z
duration_s: 282
session_id: 2ea1605c-0d43-4eaf-a765-3d6254358e74
---

# Shoplify 2.4: Admin order creation - Result

## Confirm the starting orders list ✓ passed (24.7s)
md5: c43aacb9be59680c1d7b598d24292982
Verify the Orders tab shows "Orders (2)" listing order "#1001" and order "#1002".

## Open the draft order builder ✓ passed (26.4s)
md5: 2fc6d4ae7955cadd253a4a39ca631837
Click "Draft orders" and verify a "Create draft order" card is shown with a "Customer name" field.

## Build the draft ✓ passed (141.8s)
md5: 6e3842256d1b0f35a29c38af78d3d325
Type "Marco Oduya" into "Customer name", "marco@example.com" into "Customer email", select "Canvas Market Tote" as the Product, type "2" into Quantity, click "Add line item", and verify a draft total of "$52.00" is shown.

## Save the draft ✓ passed (35.3s)
md5: a1324382351d7fa70e8395eafae9f81c
Click "Save draft order" and verify the draft orders list shows draft "D1" for "Marco Oduya" with status "open".

## Mark the draft paid ✓ passed (51.5s)
md5: 3cf14160157131cede3968d7bb7782b2
Click "Mark as paid" on draft "D1" and verify the Orders tab shows order "#1003" for "Marco Oduya" with channel "Draft order" and status "paid".

A step that takes half the run is a signal, not a problem. If that number climbed on a later run, the form is where to look, and the per step timings make that visible without any extra instrumentation.

Reading the evidence

The outcome is coverage on the tooling your own staff depend on, written in language the support team can verify. Admin regressions stop being something a merchant discovers on a Monday morning.

Internal tools are also where AI coding agents do the most unattended work, which makes an automated browser check the natural pairing. Verifying agent-written code with Claude Code covers that loop, and the three Kane CLI modes explains when to use the interactive session instead of a one-shot run.