Daily marketplace check

One answer to "what needs my attention today?" across every connected channel. Every call here is a read. Do not ship, accept, reply or change anything during the check: list what is due and offer the next step.

Load the backlane skill first if you have not checked the connection yet. Use search for the exact schema before the first call to an operation.

Before the first call

• Connections. Call list_integrations. Check every connection with status: "active" unless the user named channels. Skip disabled ones and say so. If a connection is a sandbox or playground, label its section "test data", because nothing there is a real customer.

• One script per platform. Run the reads for a platform in one execute call with Promise.all, and wrap each request so one failure does not hide the rest:

  const call = async (name, req, pick) => {
    try { return [name, pick(await codemode.request(req))]; }
    catch (e) { return [name, { error: String(e.message ?? e).slice(0, 200) }]; }
  };

• Return counts and the few oldest items, not raw responses. Responses are capped at ~6,000 tokens. Loop over pages in code when you need totals.
• Dates. Kaufland *_iso filters reject milliseconds: send 2026-11-01T06:00:00Z, not ...06:00:00.000Z (HTTP 400 "Timestamp must be provided in ISO 8601 format").

What to check

Per platform:

• OTTO. Orders are cursor-paginated (limit up to 256). Follow the next link in links and pass only its nextcursor (plus limit). An order has positionItems[], each with its own fulfillmentStatus; count items, not orders, and use mode: "AT_LEAST_ONE" if mixed orders matter. Overdue means expectedDeliveryDate is today or past and the item is still PROCESSABLE. /v3/returns answers { positionItems: [...] } and pages with page. Quantities come as resources.variations[] (sku, quantity), paged via links.
• Kaufland. Pass storefront if the seller has several. Lists return pagination.total: use limit: 1 when you only need the count. Overdue means delivery_time_expires_iso is past on a need_to_be_sent unit. Tickets: put is_seller_responsible: true on top.
• Mirakl. OR11 pages with max (≤ 100) and offset up to total_count. Overdue: shipping_deadline past on SHIPPING. Acceptance due: acceptance_decision_date on WAITING_ACCEPTANCE; missing it auto-refuses the order and hurts the shop's rating, so list these first. RT11 and M11 page with limit + page_token. A thread needs a reply when its last message is from CUSTOMER or OPERATOR (with_messages: true).
• Optional seller KPIs (Mirakl). A21 GET /api/account/statistics with periods: "LAST_MONTH" gives late_shipment_rate, incident_rate, cancelation_rate, grade and similar. It is limited to once per day: call it only if the user wants KPIs and not more than once per session. null values mean too little data, not zero.

Known test-environment gaps

• Kaufland playground: /v2/tickets and /v2/reports return 404 "not part of the playground". Report "not available in the playground" and continue.
• OTTO sandbox: /v1/return-shipments returns 503 on every call. Use /v3/returns instead.
• 401/403 on a Mirakl operation means this marketplace does not offer it to the key. Say so instead of guessing.

Report

Lead with what has a deadline. Keep it short enough to read on a phone.

1. Due today, oldest first: acceptance deadlines (Mirakl), overdue and due shipments (all), returns waiting for a decision. One line each with channel, order or return ID, age and deadline.
2. Waiting for a reply: tickets, threads and incidents with count and the oldest age. Flag anything older than 24 hours.
3. Stock: offers at 0 per channel with count; list SKUs only if there are few.
4. All clear channels in one line ("Kaufland: nothing due").
5. Which connection each section came from, and which were test data or failed.
6. Next steps as offers, each handled by its skill:
   • replies, incidents, returns: customer-service
   • not-live or rejected products: listing-health
   • ads below target: otto-ads
   • sales trends and cancellations: sales-analysis

Weekly mode

When the user asks for a weekly summary, add the last 7 days against the 7 days before, per channel: orders, units, gross revenue, cancellations and returns. Use the sales-analysis skill for the numbers and keep the daily sections above for what is still open.

Common mistakes