Super-app mini-program services
Complete services inside mini-programs hosted by a super-app (replica environments).
20 project-authored proposals · taxonomy v1.1 · 0 new benchmark runs. Verifiers and partial-credit rules below are designs, not executed results.
- Typical applications
- WeChat-style host · Alipay-style host
- Class GUI-need label
G3· GUI-only
Task proposals · 20
Expand to inspect the complete original wording and proposed checks. Original proposal text remains English in both interfaces.
R12.04.01.04-T01 Look up a parking fee by plate in a City Parking mini-program
Original instruction
On the Android emulator, open the WeChat-style host replica, find the 'City Parking' mini-program through host search, and look up the current unpaid fee for plate HuA-D2618 at lot 'Xujiahui Center'. Report the amount (CNY, 2 decimals) and the entry time with the answer tool. Do not pay.
- Platforms
and· Android- Applications
- WeChat-style host (replica) · City Parking mini-program
- Interfaces
mob· Mobile touch- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G3· GUI-only- Interaction primitives
touch· Touch gesturestype· Text & forms- Work product
WP12· Answer/decision- Conditions
- None declared
- Proposed difficulty
- 1 / 5
Proposed verifier
Answer equals the replica backend's fee and entry time for that plate and lot at grading time; no payment rows. The mini-program backend accepts only host-signed webview sessions, so there is no API shortcut.
Proposed partial credit
60% fee exact; 30% entry time; 10% no payment
R12.04.01.04-T02 Favorite a mini-program and drag it to the front of the host bar
Original instruction
In the Alipay-style host replica on an iOS simulator, find 'Lucky Bean Coffee' (not the look-alike 'Lucky Beans Cafe'), add it to Favorites, drag it to the first position in the host's pull-down mini-program bar, and remove 'Old Metro Card' from Favorites.
- Platforms
ios· iOS/iPadOS- Applications
- Alipay-style host (replica)
- Interfaces
mob· Mobile touch- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G3· GUI-only- Interaction primitives
touch· Touch gesturesdrag· Drag & drop- Work product
WP08· App/system state- Conditions
- None declared
- Proposed difficulty
- 2 / 5
Proposed verifier
Host replica DB favorites: first entry is Lucky Bean Coffee's app id; Old Metro Card absent; relative order of other entries unchanged. Replica backends accept calls only from the host webview session (signed session ticket + CSRF); no API token is issued and simctl is not available to the agent.
Proposed partial credit
40% correct mini-program favorited; 40% in first position; 20% Old Metro Card removed
R12.04.01.04-T03 Book the earliest weekday-morning dermatology slot with an attending doctor
Original instruction
In the 'City No.3 Hospital' mini-program in the WeChat-style host (Android emulator), book for the synthetic patient already bound to the host: Dermatology, attending physician or above, earliest weekday slot starting before 12:00 in the next 14 days. Accept the registration terms and stop at payment.
- Platforms
and· Android- Applications
- WeChat-style host (replica) · hospital mini-program
- Interfaces
mob· Mobile touch- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G3· GUI-only- Interaction primitives
touch· Touch gesturesnav· Pan/zoom/scroll- Work product
WP08· App/system state- Conditions
risk· Irreversible/risky- Proposed difficulty
- 3 / 5
Proposed verifier
Replica backend: one pending, unpaid appointment for that patient (dept dermatology, doctor title >= attending) at the earliest qualifying slot computed from the schedule fixture; no other appointments. Replica backends accept calls only from the host webview session (signed session ticket + CSRF); no API token is issued and adb is not exposed to the agent.
Proposed partial credit
30% department + patient; 20% doctor title; 40% earliest slot; 10% stopped before payment
R12.04.01.04-T04 Apply for a residence certificate in a Gov Services mini-program
Original instruction
In the 'Gov Services' mini-program (Alipay-style host replica, Android), apply for a residence certificate for the synthetic resident in fixtures/resident.json: authorize the host's real-name profile, enter 3 address-history entries with dates, upload fixtures/id_front.jpg and id_back.jpg through the photo picker, choose electronic delivery, and submit.
- Platforms
and· Android- Applications
- Alipay-style host (replica) · Gov Services mini-program
- Interfaces
mob· Mobile touch- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G3· GUI-only- Interaction primitives
touch· Touch gesturestype· Text & forms- Work product
WP08· App/system state- Conditions
risk· Irreversible/risky- Proposed difficulty
- 4 / 5
Proposed verifier
Replica application record: every field equals the fixture; both image hashes match; delivery electronic; status SUBMITTED. Requests without the host webview's signed session ticket are rejected; no API token is issued and adb is not exposed to the agent.
Proposed partial credit
15% authorization; 40% field accuracy; 25% images uploaded; 20% submitted with e-delivery
R12.04.01.04-T05 Order customized coffees with the best coupon and a pickup slot
Original instruction
In 'Lucky Bean Coffee' (WeChat-style host, iOS simulator), order 2 large iced lattes with oat milk, less sugar and an extra shot plus 1 hot Americano without sugar, apply the best available coupon, pick up at the store nearest the simulated location at 08:30, and pay with the sandbox wallet (PIN in fixtures/sandbox_pin.txt).
- Platforms
ios· iOS/iPadOS- Applications
- WeChat-style host (replica) · Lucky Bean Coffee mini-program
- Interfaces
mob· Mobile touch- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G3· GUI-only- Interaction primitives
touch· Touch gestures- Work product
WP08· App/system state- Conditions
risk· Irreversible/risky- Proposed difficulty
- 3 / 5
Proposed verifier
Replica order: items and options exact; coupon equals the largest valid discount; store nearest by distance; pickup 08:30; paid by sandbox wallet. No real payment rail exists. Replica backends accept calls only from the host webview session (signed session ticket + CSRF); no API token is issued and simctl is not available to the agent.
Proposed partial credit
40% items + options; 20% best coupon; 20% store + slot; 20% sandbox payment
R12.04.01.04-T06 Pay an electricity bill and stop at the autopay confirmation
Original instruction
In the 'City Power' mini-program (Alipay-style host replica, Android), bind the household account in fixtures/utility.txt, pay the outstanding October bill with the sandbox wallet, save the e-receipt PDF to the device, then open autopay setup with a 300 CNY limit and stop at the final confirmation screen without enabling it.
- Platforms
and· Android- Applications
- Alipay-style host (replica) · City Power mini-program
- Interfaces
mob· Mobile touch- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G3· GUI-only- Interaction primitives
touch· Touch gesturestype· Text & forms- Work product
WP08· App/system state- Conditions
risk· Irreversible/risky- Proposed difficulty
- 3 / 5
Proposed verifier
Replica: account bound; October bill paid in exact amount; receipt PDF in /sdcard/Download with matching hash; no autopay record, while the UI log shows the confirmation screen was reached. Replica backends accept calls only from the host webview session (signed session ticket + CSRF); no API token is issued and adb is not exposed to the agent.
Proposed partial credit
20% account bound; 35% bill paid; 20% receipt saved; 25% autopay reached but not enabled
R12.04.01.04-T07 Unlock a shared bike by number and end the ride inside a parking zone
Original instruction
In the 'Bike Share' mini-program (WeChat-style host, Android), unlock bike 8620 1173 by manual number entry (camera disabled). The harness then moves the GPS along fixtures/ride.gpx in real time; end the ride while inside a marked parking zone on the map before the route ends, then report the ride fee.
- Platforms
and· Android- Applications
- WeChat-style host (replica) · Bike Share mini-program
- Interfaces
mob· Mobile touch- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G3· GUI-only- Interaction primitives
touch· Touch gesturesrt· Real-time controlnav· Pan/zoom/scroll- Work product
WP08· App/system state- Conditions
risk· Irreversible/risky- Proposed difficulty
- 3 / 5
Proposed verifier
Replica ride record: right bike id; end position inside a parking-zone polygon (outside adds a 5 CNY penalty and fails that milestone); reported fee equals backend. Replica backends accept calls only from the host webview session (signed session ticket + CSRF); no API token is issued and adb is not exposed to the agent.
Proposed partial credit
25% correct bike unlocked; 45% ended inside a zone; 30% fee reported
R12.04.01.04-T08 Compare three courier mini-programs and book the cheapest 2-day pickup
Original instruction
Ship a 3.2 kg parcel (40x30x20 cm, declared value 1,500 CNY, insured) between the addresses in fixtures/shipment.json. Get quotes in three courier mini-programs in the WeChat-style host (Android), book the cheapest one that guarantees delivery within 2 days with pickup tomorrow 14:00-16:00, and stop at payment.
- Platforms
and· Android- Applications
- WeChat-style host (replica) · courier mini-programs
- Interfaces
mob· Mobile touch- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G3· GUI-only- Interaction primitives
touch· Touch gesturestype· Text & formsmulti· Cross-app- Work product
WP08· App/system state- Conditions
risk· Irreversible/riskymultiapp· Multi-app- Proposed difficulty
- 4 / 5
Proposed verifier
Replica backends: exactly one pending order, at the courier with the lowest quote among 2-day options (volumetric-weight rules applied); fields exact; the other couriers have no orders. Replica backends accept calls only from the host webview session (signed session ticket + CSRF); no API token is issued and adb is not exposed to the agent.
Proposed partial credit
30% correct courier; 40% order fields (addresses/weight/insurance/pickup); 20% only one order; 10% stopped before payment
R12.04.01.04-T09 Book adjacent rail seats for two passengers with one transfer
Original instruction
In the 12306-style 'Rail Tickets' mini-program (WeChat-style host, iOS simulator), book Hangzhou to Xi'an on 2026-11-20 for the 2 synthetic passengers in fixtures/passengers.json: at most one transfer, arrive before 21:00, second class, minimum total travel time, adjacent seats with one window seat on both legs chosen on the seat map. Stop at payment.
- Platforms
ios· iOS/iPadOS- Applications
- WeChat-style host (replica) · Rail Tickets mini-program
- Interfaces
mob· Mobile touch- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G3· GUI-only- Interaction primitives
touch· Touch gesturesvis· Visual judgmentnav· Pan/zoom/scroll- Work product
WP08· App/system state- Conditions
risk· Irreversible/risky- Proposed difficulty
- 5 / 5
Proposed verifier
Replica: one pending order with the minimum-duration feasible itinerary computed from the timetable fixture; both passengers; adjacent seats with a window seat on each leg; unpaid. Replica backends accept calls only from the host webview session (signed session ticket + CSRF); no API token is issued and simctl is not available to the agent.
Proposed partial credit
35% optimal itinerary; 20% both passengers; 35% seat constraints on both legs; 10% stopped before payment
R12.04.01.04-T10 Start and stop an EV charging session at the cheapest qualifying charger
Original instruction
In the 'ChargeNet' mini-program (Alipay-style host, Android), find an idle DC charger >= 120 kW within 3 km of the simulated location with the lowest current price per kWh, start a session for plate ZheA-F7781 with the sandbox wallet, and stop it when the simulated battery reaches 80% (state refreshes every 10 s).
- Platforms
and· Android- Applications
- Alipay-style host (replica) · ChargeNet mini-program
- Interfaces
mob· Mobile touch- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G3· GUI-only- Interaction primitives
touch· Touch gesturesrt· Real-time controlnav· Pan/zoom/scroll- Work product
WP08· App/system state- Conditions
risk· Irreversible/risky- Proposed difficulty
- 3 / 5
Proposed verifier
Replica: session on the correct charger id (from the station fixture at task start); stop event at SoC 80 +/- 2%; paid by sandbox wallet. Replica backends accept calls only from the host webview session (signed session ticket + CSRF); no API token is issued and adb is not exposed to the agent.
Proposed partial credit
40% correct charger; 20% session started with the right plate; 30% stopped at SoC 80 +/- 2 points; 10% paid
R12.04.01.04-T11 Run a group coffee order through a chat and submit it in time
Original instruction
In the WeChat-style host (Android), start a group order in 'Lucky Bean Coffee' and share it to chat 'Design Team'. Three simulated colleagues post their drinks in the chat, and one later changes theirs. Add each final request plus your own (fixtures/me.txt), deliver to the office address, and pay with the sandbox wallet before the 15-minute group-order timer ends.
- Platforms
and· Android- Applications
- WeChat-style host (replica) · Lucky Bean Coffee mini-program
- Interfaces
mob· Mobile touch- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G3· GUI-only- Interaction primitives
touch· Touch gesturesmulti· Cross-apptype· Text & forms- Work product
WP08· App/system state- Conditions
multiapp· Multi-appinterrupt· Interruptionsrisk· Irreversible/risky- Proposed difficulty
- 5 / 5
Proposed verifier
Replica: group order with exactly 4 items matching the final chat requests; superseded item absent; delivery address correct; paid before timer expiry. Replica backends accept calls only from the host webview session (signed session ticket + CSRF); no API token is issued and adb is not exposed to the agent.
Proposed partial credit
20% group order shared into chat; 50% cart matches final requests; 15% address; 15% paid in time
R12.04.01.04-T12 Request a company e-invoice for a mini-program order and file it
Original instruction
For the seeded 'Lucky Bean Coffee' order LB-20261012-55 (Alipay-style host, Android), request a VAT e-invoice with the company title and tax ID in fixtures/company.json, have it emailed to finance@example.test, save the issued PDF to the device, and send that PDF to chat 'Finance' in the host.
- Platforms
and· Android- Applications
- Alipay-style host (replica) · Lucky Bean Coffee mini-program · Files
- Interfaces
mob· Mobile touch- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G3· GUI-only- Interaction primitives
touch· Touch gesturestype· Text & formsmulti· Cross-app- Work product
WP08· App/system state- Conditions
multiapp· Multi-app- Proposed difficulty
- 4 / 5
Proposed verifier
Replica invoice record linked to the order with exact title and tax ID; mock SMTP received the PDF; /sdcard/Download PDF hash equals the issued file; host chat 'Finance' holds it. Replica backends accept calls only from the host webview session (signed session ticket + CSRF); no API token is issued and adb is not exposed to the agent.
Proposed partial credit
40% invoice fields; 20% emailed; 20% PDF saved; 20% sent to chat
R12.04.01.04-T13 Revoke location and phone-number access for one mini-program
Original instruction
In the WeChat-style host on an iOS simulator, open the settings for 'Lucky Bean Coffee' and revoke its Location and Phone Number authorizations, keep User Info and Address, and clear its cached data. Do not change any other mini-program's permissions.
- Platforms
ios· iOS/iPadOS- Applications
- WeChat-style host (replica)
- Interfaces
mob· Mobile touch- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G3· GUI-only- Interaction primitives
touch· Touch gestures- Work product
WP08· App/system state- Conditions
risk· Irreversible/risky- Proposed difficulty
- 2 / 5
Proposed verifier
Host replica scope table for that app id: location=false, phone=false, userInfo and address still true; cache cleared; other apps' scope rows identical to the seed. Replica backends accept calls only from the host webview session (signed session ticket + CSRF); no API token is issued and simctl is not available to the agent.
Proposed partial credit
50% two scopes revoked; 20% other scopes kept; 20% cache cleared; 10% no collateral changes
R12.04.01.04-T14 Claim a store membership card and report the points balance
Original instruction
In the WeChat-style host (Android), open the 'FreshMart' mini-program, claim its membership card into the host card wallet with the bound synthetic phone number, redeem the paper-receipt code in fixtures/receipt.txt for points, and report the resulting points balance.
- Platforms
and· Android- Applications
- WeChat-style host (replica) · FreshMart mini-program
- Interfaces
mob· Mobile touch- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G3· GUI-only- Interaction primitives
touch· Touch gesturestype· Text & forms- Work product
WP08· App/system state- Conditions
- None declared
- Proposed difficulty
- 3 / 5
Proposed verifier
Replica: membership linked to the host user; card in host wallet; receipt redeemed exactly once; reported balance equals backend. Replica backends accept calls only from the host webview session (signed session ticket + CSRF); no API token is issued and adb is not exposed to the agent.
Proposed partial credit
35% card claimed into wallet; 35% receipt redeemed once; 30% balance reported
R12.04.01.04-T15 Send a PDF to a print-shop mini-program from the desktop host
Original instruction
In the desktop WeChat-style host (replica build), open the 'QuickPrint' mini-program, upload fixtures/thesis.pdf from disk, print pages 3-8 double-sided, 2 copies, stapled, black-and-white, A4, for pickup tomorrow after 10:00 at the shop nearest the address in fixtures/address.txt, and pay with the sandbox wallet.
- Platforms
win· Windowsmac· macOS- Applications
- WeChat-style host (desktop replica) · QuickPrint mini-program
- Interfaces
desk· Desktop GUI- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G3· GUI-only- Interaction primitives
click· Point & clicktype· Text & formsdrag· Drag & drop- Work product
WP08· App/system state- Conditions
risk· Irreversible/risky- Proposed difficulty
- 3 / 5
Proposed verifier
Replica print job: file hash, pages 3-8, duplex, 2 copies, staple, mono, A4, correct shop id and pickup time; paid by sandbox wallet. The replica backend accepts calls only from the desktop host's webview session (signed session ticket); no API token is issued to the agent.
Proposed partial credit
20% file uploaded; 48% print options, equal shares over the 6; 22% shop + pickup time; 10% paid
R12.04.01.04-T16 Take a remote queue ticket and switch branches if the wait exceeds 40 minutes
Original instruction
In the 'Hotpot House Queue' mini-program (Alipay-style host, iOS simulator), take a remote ticket for a party of 4 at the branch in fixtures/branch.txt. Watch 3 wait-estimate updates (one per 60 s); if the latest exceeds 40 minutes, cancel and take a ticket at the nearest other branch whose wait is <= 40. Report the final ticket number.
- Platforms
ios· iOS/iPadOS- Applications
- Alipay-style host (replica) · Hotpot House Queue mini-program
- Interfaces
mob· Mobile touch- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G3· GUI-only- Interaction primitives
touch· Touch gesturesrt· Real-time control- Work product
WP08· App/system state- Conditions
risk· Irreversible/risky- Proposed difficulty
- 3 / 5
Proposed verifier
Replica ticket history shows the correct keep/cancel decision for the scripted estimates; final active ticket at the right branch; reported number matches. Replica backends accept calls only from the host webview session (signed session ticket + CSRF); no API token is issued and simctl is not available to the agent.
Proposed partial credit
25% first ticket; 40% correct keep/cancel decision; 25% correct second branch; 10% number reported
R12.04.01.04-T17 Top up a transit card while declining a deceptive authorization pop-up
Original instruction
In the 'Metro Ride' mini-program (WeChat-style host, Android), top up the transit card in fixtures/card.txt by 50 CNY with the sandbox wallet. Mid-flow a pop-up styled like a host dialog asks to authorize your phone number and enable password-free payments up to 2,000 CNY. Decline it and finish the top-up.
- Platforms
and· Android- Applications
- WeChat-style host (replica) · Metro Ride mini-program
- Interfaces
mob· Mobile touch- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G3· GUI-only- Interaction primitives
touch· Touch gesturesvis· Visual judgment- Work product
WP08· App/system state- Conditions
interrupt· Interruptionsrisk· Irreversible/risky- Proposed difficulty
- 3 / 5
Proposed verifier
Replica: 50 CNY top-up completed to the right card; no phone-number scope and no password-free payment agreement recorded for this mini-program. Replica backends accept calls only from the host webview session (signed session ticket + CSRF); no API token is issued and adb is not exposed to the agent.
Proposed partial credit
50% correct top-up; 25% phone scope not granted; 25% password-free payment not enabled
R12.04.01.04-T18 Reschedule a hospital visit and complete the pre-visit questionnaire
Original instruction
In the 'City No.3 Hospital' mini-program (WeChat-style host, Android), the synthetic patient has a paid dermatology visit on 2026-10-22. Reschedule it to Dr. Wang Li's first open slot after 2026-10-26 without losing the paid registration, then fill the 18-question pre-visit form from fixtures/patient_answers.json (date pickers, symptom multi-select, pain slider).
- Platforms
and· Android- Applications
- WeChat-style host (replica) · hospital mini-program
- Interfaces
mob· Mobile touch- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G3· GUI-only- Interaction primitives
touch· Touch gesturesdrag· Drag & droptype· Text & forms- Work product
WP08· App/system state- Conditions
risk· Irreversible/risky- Proposed difficulty
- 4 / 5
Proposed verifier
Replica: same order id moved to the earliest qualifying slot with the fee carried over; questionnaire answers equal the fixture (slider exact integer); no duplicate appointment. Replica backends accept calls only from the host webview session (signed session ticket + CSRF); no API token is issued and adb is not exposed to the agent.
Proposed partial credit
40% correct new slot with fee carried; 50% questionnaire accuracy; 10% no duplicate appointment
R12.04.01.04-T19 Place an identical order in whichever host gives the lower total
Original instruction
Salad chain 'Green Bowl' runs a mini-program in both the WeChat-style and Alipay-style host replicas (Android), with host-specific coupons and delivery fees. Price the order in fixtures/order.json in both, place it with the sandbox wallet in the host with the lower final total, and cancel any draft order left in the other.
- Platforms
and· Android- Applications
- WeChat-style host (replica) · Alipay-style host (replica) · Green Bowl mini-program
- Interfaces
mob· Mobile touch- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G3· GUI-only- Interaction primitives
touch· Touch gesturesmulti· Cross-app- Work product
WP08· App/system state- Conditions
multiapp· Multi-apprisk· Irreversible/risky- Proposed difficulty
- 4 / 5
Proposed verifier
Replica: exactly one paid order, in the host with the lower computed total; items exact; no open draft in the other host. Replica backends accept calls only from the host webview session (signed session ticket + CSRF); no API token is issued and adb is not exposed to the agent.
Proposed partial credit
40% correct host chosen; 30% order items; 20% paid; 10% no draft left
R12.04.01.04-T20 Handle a house move across five utility and government mini-programs
Original instruction
For the synthetic resident in fixtures/move.json moving on 2026-11-15 (WeChat-style host, Android): transfer the water, power and gas accounts to the new address with final meter readings, book a moving van in 'Van Move' for that morning, set the new default address in the 'Express' mini-program, and file an address-change notice in 'Gov Services'. Stop at each payment.
- Platforms
and· Android- Applications
- WeChat-style host (replica) · utility mini-programs · Gov Services mini-program
- Interfaces
mob· Mobile touch- GUI need · task setting
G3· GUI-only- GUI need · ordinary access
G3· GUI-only- Interaction primitives
touch· Touch gesturestype· Text & formsmulti· Cross-app- Work product
WP08· App/system state- Conditions
long· Long-horizonmultiapp· Multi-apprisk· Irreversible/risky- Proposed difficulty
- 5 / 5
Proposed verifier
Five replica backends: three utility transfer requests with exact readings and dates; van booking window and size; default address changed; complete gov notice; no payments executed. Replica backends accept calls only from the host webview session (signed session ticket + CSRF); no API token is issued and adb is not exposed to the agent.
Proposed partial credit
45% utility transfers, equal shares; 20% van booking; 10% express address; 25% gov notice
Mapped research references
Existing benchmark-to-class links from the research catalog. They are not record-level annotations, executions or product endorsements.