Drivethru Adidas Click

Browser-driven adidas Click B2B ordering toolkit — places purchase orders on the adidas Click portal with Playwright, since adidas exposes no public ordering API. Accepts style number, color, size, quantity, a purchase-order number, and a ship-to address, drives the cart/checkout, and (on confirm) submits the order. Use whenever the user needs to place or draft a purchase order on adidas Click.

adidas Click B2B toolkit

adidas Click is a B2B ordering portal with no public API, so this skill places purchase orders by driving the portal with Playwright. Every tool is reached through one CLI entrypoint:

bash

echo '<json-args>' | python3 scripts/adidas.py <action>

The action is the first CLI argument; arguments are a JSON object on stdin. Each call prints a single JSON object on stdout, or {"error": {"type": ..., "message": ...}} with a non-zero exit code on failure.

Playwright is imported lazily, so the skill loads without it. The Playwright package is a declared uv dependency; its Chromium browser binary (a separate ~150 MB download that pip/uv doesn't fetch, and for which OpenClaw has no install-time hook) is installed automatically on the first run if missing — so no manual python -m playwright install chromium is needed. That first run therefore includes a one-time browser download. If the auto-install fails (e.g. no network, or missing system libraries), the tool returns a clear config_error pointing at the manual command.

Build status — full order flow implemented. The skill drives a purchase order end to end: login → add line quantities to the active cart → checkout (Customer PO, delivery location — default / saved / one-time dropship — and shipping method) → Next → Calc. Net Price (waits for the "Done!" tell so the wholesale discount always applies) → Order Now → confirmation number parsed from the redirect URL. confirm: false is a full dry run; confirm: true places a real order (no sandbox). The assembled flow has been built and unit-checked selector-by-selector but not yet run against the live site in one pass — watch the first real end-to-end run. See references/order_flow_notes.md for the step map and captured selectors.

When to use this skill

Reach for create-purchase-order when the request involves placing (or drafting) a purchase order on adidas Click: given a style number, color, size, quantity, a PO number, and a ship-to address, put the order into the portal cart/checkout and — on explicit confirmation — submit it.

Do not use it for other footwear/apparel vendors, and never invent portal selectors from prose — the flow lives in the captured references/order_flow_notes.md.

Actions

Action	Risk	stdin JSON (key fields)
create-purchase-order	high — portal write (browser)	{purchase_order: {po_number, lines: [{style, size, quantity}], ship_to?, delivery_location_id?, ship_method?, on_insufficient_stock?, spread_delivery?}, confirm}

Run python3 scripts/adidas.py with no action to print the action list.

create-purchase-order input

The order can be passed nested under purchase_order or as top-level fields:

json

{
  "purchase_order": {
    "po_number": "PO-12345",
    "lines": [
      {"style": "JW4306", "size": "M", "quantity": 24},
      {"style": "JW4306", "size": "L", "quantity": 12}
    ],
    "ship_to": null,
    "delivery_location_id": null,
    "ship_method": null,
    "spread_delivery": false,
    "notes": null
  },
  "confirm": false
}

Each line is {style, size, quantity}. color is optional — adidas article numbers encode the color (e.g. JW4306 = the black colorway), so navigating to the article lands on the right color with no picker.

Delivery location precedence: delivery_location_id (pick a saved location) > ship_to (add a one-time / dropship location) > default (leave the cart's preset). A one-time ship_to.state accepts a 2-letter abbreviation or the full name. ship_method is left at the cart default unless set, and accepts a FedEx service code (FDGP, FEDE, FED4, …) or its exact label ("FedEx Ground"). po_number is capped at 18 characters and may contain only letters, numbers, space, and / _ . ? &; any other character (e.g. a hyphen) is auto-replaced with _ and the substitution is reported in the result's warnings.

By default (new_cart: true) the run creates its own fresh cart (named with the PO) right after login and makes it active, so on a shared account it never adds to — or checks out — a teammate's cart. If a cart with the same PO name already exists (e.g. a leftover from an errored prior run) it is deleted first so the run never piles onto stale quantities — only carts with that exact PO name are removed, and any deletion is noted in warnings. Pass new_cart: false to use the current active cart instead.

Pass screenshot_path to capture the filled page for review.

Out-of-stock handling (read this)

Before entering quantities, the tool reads each size's availability (scrolling the horizontal size row as needed to load off-screen sizes). A size is either backorderable (stock below the requested quantity, incl. 0 with a restock date) or not available (the "X" cell — will never be restocked, cannot be ordered at all; it can only be removed or substituted). When a line's full quantity is not available, behavior depends on on_insufficient_stock:

• pause (default) — nothing is ordered (even with confirm: true). The result comes back status: "needs_confirmation" with an out_of_stock list ({style, size, requested, available}) and a message. The agent must stop and confirm with the user, then re-run.
• order — order the short line(s) anyway, accepting delayed delivery (adidas spreads the backordered portion over future dates).
• skip — remove the out-of-stock line(s) and order the rest.

Agent guidance: if the user's request pre-authorizes a choice, set the flag and skip the pause — e.g. "go ahead and order anything out of stock" → on_insufficient_stock: "order"; "remove any out-of-stock items without asking" → on_insufficient_stock: "skip". Otherwise leave it at pause; on a needs_confirmation result, message the user with the out_of_stock details and the three choices, wait for their answer, then resume:

1. Order them anyway (delayed delivery) → re-run with on_insufficient_stock: "order".
2. Don't order them → re-run with on_insufficient_stock: "skip".
3. Substitute (e.g. a different size/style so items match) → edit the lines accordingly and re-run (optionally with a policy for any still-short substitutes).

spread_delivery is the lower-level knob behind order: on a per-line spread prompt, false (default) declines (single delivery), true accepts.

The result reports total_quantity (summed pieces, on dry runs too) and, on a placed order (confirm: true), each line's net line_total and unit_price (net ÷ quantity) plus the order order_total (net wholesale), read from the priced review page.

Credentials

The adidas Click portal login can be supplied three ways (later wins):

1. Environment variables (preferred for a deployed agent):

   bash

   ADIDAS_CLICK_USERNAME=...
   ADIDAS_CLICK_PASSWORD=...
   ADIDAS_CLICK_BASE_URL=https://b2bportal.adidas-group.com   # override for region host
   

2. Inline in the stdin JSON — pass username, password, and base_url alongside the tool's own arguments.

3. CLI flags — --username / --password / --base-url after the action. Handy for local runs with no env vars and no secrets in the order JSON:

   bash

   echo '{"purchase_order": {...}, "confirm": false}' \
     | python3 scripts/adidas.py create-purchase-order --username "$U" --password "$P"
   

   Command-line arguments are visible to other processes (ps) and your shell history — for sensitive/automated use, prefer env vars or the stdin JSON.

If no credentials are present via any of these, the tool exits with {"error": {"type": "config_error", ...}} (exit code 2) — treat that as a signal to ask the user for the missing fields. Never guess or reuse credentials across accounts, or paste secrets the user did not provide.

Write safety

create-purchase-order is the only (side-effecting) action.

• It requires "confirm": true to place the order. Without it, it logs in, fills the cart + checkout, and returns a dry-run preview without submitting. Confirm with the user before placing.
• Placing an order on adidas Click is a real purchase — there is no sandbox. Only run confirm: true for an order the user actually intends to place.

Error model

Failures print {"error": {...}} and exit non-zero:

• config_error (exit 2) — missing/invalid credentials or an un-captured step.
• api_error — the portal rejected an action or a page did not match expectations. Includes surface, operation, retryable.
• connection_error — network / browser-launch / navigation failure (retryable: true).
• validation_error — bad input JSON or a missing required field.
• usage / unknown_action (exit 2) — bad CLI invocation; the message lists the valid actions.

Surface the human-readable message to the user. Do not retry on config_error or validation_error.

References

• references/order_flow_notes.md — the reverse-engineered portal ordering flow (step map, captured selectors, and the capture checklist for the next step). Start here to continue the build.