Taobao/Tmall Cart Checkout Optimizer

Works out the cheapest way to buy a Taobao or Tmall cart: the total spend, whether to split it into separate checkouts, whether an extra item is worth adding, and a step-by-step purchase sequence you can follow on screen.

Scope is limited to ordinary domestic shopping on Taobao and Tmall. Other platforms, Tmall Global and cross-border tax are out of scope.

When it helps

• A cart with a dozen items, and you want to know the real total and the cheapest combination
• You have official discounts, 88VIP rates, multi-item discounts, store coupons, red packets and coins, and you are unsure what stacks and in which order
• You want to compare "split into two checkouts and pay less" against "one checkout and fewer steps"
• The first order is already paid, and you need to know how to check out the rest

What you need before starting

• An agent environment that can read and write local files and run Python 3.10+ (the core uses only the standard library)
• A browser that can sign in to Taobao/Tmall. By default the agent opens a visible browser and reads your cart; if you are not signed in it opens the login page and waits for you to scan or complete verification
• If the browser is unavailable, or you would rather not sign in, you can also send cart screenshots or text and the agent will still compute the answer

How it works

1. Read-only capture — the agent opens the browser and reads the cart, item details, discount and coupon terms, and a read-only checkout preview, producing cart-snapshot.json. Nothing is modified: no cart edits, no coupon claims, no orders, no payment. You are told before any checkout preview is entered.
2. Verify the rules — for official discounts, membership rates, multi-item discounts, coupons, red packets and coins, the base amount, per-item versus whole-order rounding, caps, skippability and the discount stage each need evidence from the page in front of you. Historical campaign thresholds and rates are never assumed.
3. Exact local computation — all amounts are computed by a local Python program, never estimated by the model. A post-coupon price is not discounted twice, and tail-payment or unknown price bases stop the calculation instead of guessing.
4. An executable plan — total spend, number of checkouts, any extra item you approved, and the order in which to place each purchase.

Every task is written to session.json inside its own task directory and survives across conversations. Saying "make the lamp 199" or "the first order is already paid" updates it, and stale results are invalidated and recomputed.

What you get

Sample case (demo, explicitly labelled demo data, not current platform pricing):

Stacking demo (demo --discounts): a 100 CNY item receives an official 15% markdown, then a 95% rate on the remaining amount, then a 10 CNY red packet, then an example points deduction, for an estimated 67.22 CNY. The example only exercises the calculation model; it is not current platform rates or exchange rules.

Supported scope

• Verified full-reduction, per-N reduction with caps, discount scope and mutual exclusion, per-account coupon counts, order splitting, explicitly approved extra items, flat store shipping and paid deposits for small carts on Taobao and Tmall
• Taobao and Tmall stores are computed together inside checkout groups confirmed as combinable; Tmall-only coupons restrict their threshold by store type
• Official markdowns, membership rate discounts, full-item and tiered discounts, per-N amount reductions, per-item or whole-order rounding, per-use and shared caps, red packet and coin balances, and conditional fixed or rate payment discounts
• Current computation limits: up to 12 items and 12 discounts; complex inputs may time out and return an explicit incomplete result

Not supported: other shopping platforms, Tmall Global and cross-border tax, buy-one-get-one and nth-item discounts, paid stored value and coupon purchase costs, cashback or voucher loops, return-driven top-ups, and add-on budget constraints.

Notes

• The program only computes and records. The optional browser adapter reads your shopping account pages in read-only mode and never submits an order or pays. Placing orders and paying are always up to you.
• Discount eligibility, rates, bases, ordering, rounding and exchange rules must all be backed by the page at the time of use; they are never inferred from a discount's name.
• Page structure and selectors change as the platform is updated; behaviour on a real account depends on the live page.
• When price, shipping or a key condition is missing, the assistant keeps what it has and asks you, rather than guessing numbers.