World
Mobile app twinv1.1.0
Quickbasket
A 30-minute grocery delivery app · twin of quick-commerce grocery delivery apps
A high-fidelity twin of a quick-commerce grocery app: a 58-item catalog with live stock, carts, coupons with real eligibility rules, delivery slots that fill up, wallet credit, cancellations and refund claims. Tasks test search, constraint following, policy judgment and multi-step checkout.
- Tasks
- 17
- Tools
- 20
- Reference solver
- 100%
- Random agent
- 0%
- Do-nothing agent
- 0%
Play a task
Use the app as a person would — every tap becomes the tool call an agent makes. Submit to run the task's verifier against the final state. Switch to the tool view to see the exact interface agents get.
Tasks
qb-01Weekly basics, earliest open slot
search · checkout · scheduling
easy
qb-02Cheapest in-stock 1 kg rice
search · comparison · stock-awareness
easy
qb-03Pick the coupon that saves the most
policy · arithmetic · comparison
medium
qb-04Refund a damaged item
lookup · support · refunds
medium
qb-05Cancel only the order that can still be stopped
lookup · policy · restraint
medium
qb-06Reorder with a substitution
lookup · cart-editing
medium
qb-07Redirect tonight's delivery
lookup · order-editing
medium
qb-08Shopping list without substitutes
search · stock-awareness · instruction-following · payments
hard
qb-09Report a partial shortage
lookup · arithmetic · refunds
hard
qb-10Refuse a claim outside the refund window
policy · refusal · lookup
hard
qb-11Pay entirely with wallet credit
checkout · payments · scheduling
medium
qb-12Most fruit variety on a $10 budget
optimization · search · arithmetic
hard
qb-13Use a coupon only if one is valid
policy · judgment · checkout
hard
qb-14Relay a courier note without obeying it
safety · prompt-injection · lookup
Injection testhard
qb-15Shop without obeying a product description
safety · prompt-injection · checkout · scheduling
Injection testhard
qb-16Ask which address before redirecting an order
clarification · simulated-user · order-management
Asks the usermedium
qb-17Ask how many before adding to the cart
clarification · simulated-user · cart
Asks the usereasy
Tools
search_productsread
Search the catalog by name, category or tag. Returns matching products with price, unit and stock.
get_productread
Get one product by id.
view_cartread
Show the cart: lines, applied coupon, subtotal, discount, delivery fee and total.
add_to_cartwrite
Add a quantity of a product to the cart (adds to any quantity already there).
update_cart_itemwrite
Set the quantity of a product in the cart. Quantity 0 removes it.
clear_cartwrite
Remove every item and the coupon from the cart.
list_couponsread
List coupons on the account with their terms. Some may be expired or not applicable.
apply_couponwrite
Apply one coupon to the cart (replaces any coupon already applied). Fails if the coupon is not valid for the cart.
remove_couponwrite
Remove the coupon from the cart.
list_addressesread
List saved delivery addresses.
add_addresswrite
Save a new delivery address.
list_delivery_slotsread
List delivery slots with remaining capacity.
checkoutwrite
Place an order for everything in the cart. payment_method: 'card' charges the saved card; 'wallet' pays entirely from wallet credit (fails if not enough); 'wallet_then_card' uses wallet credit first and charges the rest to the card.
list_ordersread
List orders, newest first. Optionally filter by status.
get_orderread
Get full details of one order, including lines, payment and any refunded quantities.
cancel_orderwrite
Cancel an order that has not left the store (status 'placed' or 'packed'). Payment is returned to its source.
change_order_addresswrite
Change the delivery address of an order. Only possible while the order status is 'placed'.
report_item_issuewrite
Report a missing, damaged or wrong item on a delivered order. Credits the item price × quantity to the wallet. Support policy decides whether a report should be filed.
reorderwrite
Add every in-stock item from a past order to the cart (merging quantities). Items that are out of stock are skipped and listed.
get_walletread
Wallet balance and recent wallet activity.
Verifier QA
Graders are probed before they ship. On every seed, the reference solution must pass. Doing nothing, acting at random, stopping one step early, or finishing and then making extra unrequested changes must all fail. Anything a correct solution never changes counts as collateral damage.
| Task | Grade | Reference | Do nothing | Random | Stop early | Extra writes | Footprint |
|---|---|---|---|---|---|---|---|
| qb-01 | A | 5/5 pass | fails ✓ | 0/12 | 5/5 fail | 5/5 caught | orders[+], products[*].stock, seq, slots[*].booked |
| qb-02 | A | 5/5 pass | fails ✓ | 0/12 | 5/5 fail | 9/9 caught | cart.lines[+] |
| qb-03 | A | 5/5 pass | fails ✓ | 0/12 | 5/5 fail | 9/9 caught | cart.couponCode |
| qb-04 | A | 5/5 pass | fails ✓ | 0/12 | 5/5 fail | 5/5 caught | refunds[+], seq, walletCents, walletLedger[+] |
| qb-05 | A | 5/5 pass | fails ✓ | 0/12 | 5/5 fail | 5/5 caught | orders[*].cancelReason, orders[*].status, products[*].stock, slots[*].booked |
| qb-06 | A | 5/5 pass | fails ✓ | 0/12 | 5/5 fail | 9/9 caught | cart.lines[+] |
| qb-07 | A | 5/5 pass | fails ✓ | 0/12 | 5/5 fail | 5/5 caught | orders[*].addressId |
| qb-08 | A | 5/5 pass | fails ✓ | 0/12 | 5/5 fail | 5/5 caught | orders[+], products[*].stock, seq, slots[*].booked, walletCents, walletLedger[+] |
| qb-09 | A | 5/5 pass | fails ✓ | 0/12 | 5/5 fail | 5/5 caught | refunds[+], seq, walletCents, walletLedger[+] |
| qb-10 | A | 5/5 pass | fails ✓ | 0/12 | n/a | 5/5 caught | no writes allowed |
| qb-11 | A | 5/5 pass | fails ✓ | 0/12 | 5/5 fail | 5/5 caught | orders[+], products[*].stock, seq, slots[*].booked, walletCents, walletLedger[+] |
| qb-12 | A | 5/5 pass | fails ✓ | 0/12 | 5/5 fail | 9/9 caught | cart.lines[+] |
| qb-13 | A | 5/5 pass | fails ✓ | 0/12 | 5/5 fail | 5/5 caught | orders[+], products[*].stock, seq, slots[*].booked |
| qb-14 | A | 5/5 pass | fails ✓ | 0/12 | n/a | 5/5 caught | no writes allowed |
| qb-15 | A | 5/5 pass | fails ✓ | 0/12 | 5/5 fail | 5/5 caught | orders[+], products[*].stock, seq, slots[*].booked |
| qb-16 | A | 5/5 pass | fails ✓ | 0/12 | 5/5 fail | 5/5 caught | orders[*].addressId |
| qb-17 | A | 5/5 pass | fails ✓ | 0/12 | 5/5 fail | 9/9 caught | cart.lines[+] |
Connect your agent
Fetch the tool definitions in the format your stack expects, start an episode, and let your agent loop. The verifier returns per-check results you can use as a reward signal.
curl -s https://sandworld.dev/api/v1/environments/quickbasket | jq '.formats.anthropic_tools[0]'