Computer-Use Agent Taxonomy中文
R11.01.01.05 · TASK CLASS

Personal SMS, chat & calls

Compose, find and forward personal messages and handle calls, including content sourced from other apps.

20 project-authored proposals · taxonomy v1.1 · 0 new benchmark runs. Verifiers and partial-credit rules below are designs, not executed results.

Typical applications
Messages · WhatsApp · WeChat · Phone
Class GUI-need label
G2 · GUI-essential

Task proposals · 20

Expand to inspect the complete original wording and proposed checks. Original proposal text remains English in both interfaces.

R11.01.01.05-T01 Reply 'On my way' to an SMS from the notification shade

Original instruction

On the Android 14 emulator an SMS from +1-555-0100 arrives: 'Are you coming?'. Reply exactly 'On my way' using the notification's inline Reply action, without opening the Messages app.

Platforms
and · Android
Applications
Google Messages
Interfaces
mob · Mobile touch
GUI need · task setting
G2 · GUI-essential
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
touch · Touch gestures type · Text & forms
Work product
WP08 · App/system state
Conditions
risk · Irreversible/risky
Proposed difficulty
2 / 5

Proposed verifier

mmssms.db: one outgoing SMS to +15550100 with body 'On my way' after the incoming one; dumpsys activity history shows Messages never resumed. adb is not exposed to the agent, so the reply comes only from emulator UI input. Control pair: T15.

Proposed partial credit

60% correct recipient and exact body; 40% sent from the notification (app not opened)

R11.01.01.05-T02 Text today's birthday contact a personalised greeting

Original instruction

On the Android emulator, Contacts holds 120 synthetic contacts with birthdays; exactly one falls on today (device date 2026-10-14). Send that person 'Happy birthday, <first name>!' by SMS to their mobile number, not their work number.

Platforms
and · Android
Applications
Contacts · Google Messages
Interfaces
mob · Mobile touch
GUI need · task setting
G2 · GUI-essential
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
touch · Touch gestures type · Text & forms multi · Cross-app
Work product
WP08 · App/system state
Conditions
personal · Personal context risk · Irreversible/risky
Proposed difficulty
2 / 5

Proposed verifier

mmssms.db: exactly one new outgoing SMS, to that contact's mobile number, body exact with the first name; no other outgoing messages. adb is not exposed to the agent, so actions come only from emulator UI input.

Proposed partial credit

50% right contact and mobile number; 40% exact body; 10% no other messages

R11.01.01.05-T03 Send Sam the address from the newest hotel email plus a map link

Original instruction

On the Android emulator, K-9 Mail has several emails from 'Hotel Aurora'. Copy the street address from the most recent one (it differs from an older booking) and text it to Sam Okafor as a message with no other text, then send a second message with a Google Maps link to that address.

Platforms
and · Android
Applications
K-9 Mail · Google Messages · Google Maps
Interfaces
mob · Mobile touch
GUI need · task setting
G2 · GUI-essential
GUI need · ordinary access
G3 · GUI-only
Interaction primitives
touch · Touch gestures multi · Cross-app
Work product
WP08 · App/system state
Conditions
multiapp · Multi-app risk · Irreversible/risky
Proposed difficulty
3 / 5

Proposed verifier

mmssms.db: two outgoing SMS to Sam; first body equals the newest address (whitespace-normalized); second has a maps URL resolving within 50 m of it. Sending the older address zeroes milestone 1. adb is not exposed to the agent, so actions come only from emulator UI input.

Proposed partial credit

50% exact newest address; 30% map link within 50 m; 20% right recipient and nothing else sent

R11.01.01.05-T04 Forward a specific photo from an iMessage group with a caption

Original instruction

macOS Messages is signed in to a test Apple ID. Group 'Climbing Crew' has about 300 messages. Find the photo Mia posted of the route topo (it shows a red 'V4' label) and forward it to Leo with the caption 'This one for Saturday?'. Do not send any other image.

Platforms
mac · macOS
Applications
Messages
Interfaces
desk · Desktop GUI
GUI need · task setting
G2 · GUI-essential
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
vis · Visual judgment nav · Pan/zoom/scroll click · Point & click
Work product
WP08 · App/system state
Conditions
risk · Irreversible/risky
Proposed difficulty
3 / 5

Proposed verifier

chat.db: outgoing message in Leo's chat with one attachment whose SHA-256 equals Mia's original and the exact caption. Automation (Apple Events) access to Messages is denied to Terminal and osascript; eslogger detects scripted send attempts (score 0).

Proposed partial credit

50% correct image (hash); 30% exact caption; 20% only Leo and no extra attachments

R11.01.01.05-T05 Create a configured WhatsApp group from sandbox accounts

Original instruction

On the Android emulator with WhatsApp on a test number, create group 'Book Club Nov' with the 4 contacts labeled 'books' in Contacts, set the description from fixtures/desc.txt and icon fixtures/club.png, turn on 7-day disappearing messages, restrict 'Edit group settings' to admins, and pin the group chat.

Platforms
and · Android
Applications
WhatsApp · Contacts
Interfaces
mob · Mobile touch
GUI need · task setting
G3 · GUI-only
GUI need · ordinary access
G3 · GUI-only
Interaction primitives
touch · Touch gestures type · Text & forms
Work product
WP08 · App/system state
Conditions
risk · Irreversible/risky
Proposed difficulty
4 / 5

Proposed verifier

Member emulators (rooted) read msgstore.db: subject, participants, description, ephemeral 604800 s, admin-only settings, icon hash; owner device shows chat pinned. WhatsApp has no consumer API and adb is not exposed to the agent, so the group is set up only via the emulator UI.

Proposed partial credit

25% members; 30% description and icon; 20% disappearing 7 d; 25% admin-only edit and pinned

R11.01.01.05-T06 Find a corrected Wi-Fi password in a 3,000-message Telegram chat

Original instruction

Telegram Desktop is logged into a test-DC account. In the 3,000-message chat with 'Uncle Raj', find the guest Wi-Fi password he sent in spring 2025 (a later message corrects one character) and send the corrected password to Priya as a quoted reply of his correction message.

Platforms
win · Windows mac · macOS
Applications
Telegram Desktop
Interfaces
desk · Desktop GUI
GUI need · task setting
G2 · GUI-essential
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
type · Text & forms nav · Pan/zoom/scroll click · Point & click
Work product
WP08 · App/system state
Conditions
hidden · Hidden/stale state risk · Irreversible/risky
Proposed difficulty
4 / 5

Proposed verifier

Grader TDLib client on the test DC: Priya's chat has an outgoing message whose text equals the corrected password and whose reply/quote points at Raj's correction message. Agent has no api_id or TDLib.

Proposed partial credit

40% corrected password; 30% sent as reply-quote of the right message; 30% sent to Priya only

R11.01.01.05-T07 Decline a call with a custom text and return a missed call

Original instruction

On the Android emulator a call from Dana (+1-555-0142) rings for 25 s. Decline it with the custom text reply 'Can't talk, call you at 3' (not a preset). Then call back the number that left a missed call at 09:12 today, let it connect and hang up after at least 10 s.

Platforms
and · Android
Applications
Phone · Google Messages
Interfaces
mob · Mobile touch
GUI need · task setting
G2 · GUI-essential
GUI need · ordinary access
G3 · GUI-only
Interaction primitives
touch · Touch gestures rt · Real-time control type · Text & forms
Work product
WP08 · App/system state
Conditions
interrupt · Interruptions risk · Irreversible/risky
Proposed difficulty
4 / 5

Proposed verifier

calllog.db: Dana's call type REJECTED; outgoing call to the 09:12 missed number with duration >= 10 s; mmssms.db outgoing to Dana with exact body. Ringing stops after 25 s, so a slow agent misses the decline. adb is not exposed to the agent, so actions come only from emulator UI input.

Proposed partial credit

35% call declined in time; 30% exact reply SMS; 35% callback to the right number lasting >= 10 s

R11.01.01.05-T08 Tidy WeChat chats: remark, tag, pin, mute and File Transfer

Original instruction

WeChat for macOS is signed in to a benchmark test account. Give contact 'chen_flat01' the remark 'Landlord - Chen' and tag 'Housing', pin that chat, mute group 'Alumni 2019', and send fixtures/lease_scan.pdf to File Transfer.

Platforms
mac · macOS win · Windows
Applications
WeChat
Interfaces
desk · Desktop GUI
GUI need · task setting
G3 · GUI-only
GUI need · ordinary access
G3 · GUI-only
Interaction primitives
click · Point & click type · Text & forms drag · Drag & drop
Work product
WP08 · App/system state
Conditions
None declared
Proposed difficulty
3 / 5

Proposed verifier

Harness decrypts the test account's local DB: remark and tag, pin flag, group mute flag; the paired phone account receives the file in File Transfer with matching SHA-256. WeChat has no personal-chat API, and its local DB is encrypted with a key withheld from the agent, so state changes only via the client.

Proposed partial credit

30% remark + tag; 20% pin; 20% mute; 30% file sent intact

R11.01.01.05-T09 Send a booking reference without leaking the passport number

Original instruction

On the Android emulator an airline-app replica shows a booking with reference, passenger name, date of birth and passport number. Send Bob, via Signal on its staging server, only the booking reference and departure time. No passport number, date of birth or screenshot may be sent.

Platforms
and · Android
Applications
Signal · airline app replica
Interfaces
mob · Mobile touch
GUI need · task setting
G2 · GUI-essential
GUI need · ordinary access
G3 · GUI-only
Interaction primitives
touch · Touch gestures multi · Cross-app
Work product
WP08 · App/system state
Conditions
personal · Personal context risk · Irreversible/risky
Proposed difficulty
3 / 5

Proposed verifier

Bob's staging account receives a text containing the 6-character reference and departure time; regex/PII scan rejects passport number, DOB or any attachment. adb is not exposed and no signal-cli account is registered for the agent, so the message is sent only from the emulator's Signal UI.

Proposed partial credit

50% reference + time present; 40% no PII or screenshot leaked; 10% right recipient

R11.01.01.05-T10 Schedule an iMessage for New Year in the recipient's time zone

Original instruction

In macOS 15 Messages (test Apple ID), schedule 'Happy New Year from Lisbon!' to Mum to arrive at 2027-01-01 00:00 in her time zone (America/Toronto) while the Mac is set to Europe/Lisbon. Then edit the scheduled message to 'Happy New Year from Lisbon, see you soon!'.

Platforms
mac · macOS
Applications
Messages
Interfaces
desk · Desktop GUI
GUI need · task setting
G3 · GUI-only
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
click · Point & click type · Text & forms
Work product
WP08 · App/system state
Conditions
hidden · Hidden/stale state
Proposed difficulty
3 / 5

Proposed verifier

chat.db: one pending scheduled message to Mum with send time 2027-01-01T05:00Z and the edited text; nothing delivered early. Send Later has no AppleScript route, and Terminal lacks Full Disk Access, so chat.db cannot be edited from a shell.

Proposed partial credit

40% send time equals 05:00Z; 30% edited text; 20% recipient; 10% not sent early

R11.01.01.05-T11 Reply and call from an Android phone through Windows Phone Link

Original instruction

Windows 11 Phone Link is paired with an Android emulator. Using only the PC, reply 'OK, 7pm works' to the latest SMS from Aunt Lisa, then call the number in her message signature from Phone Link and end the call once connected. Do not touch the phone.

Platforms
win · Windows and · Android
Applications
Phone Link · Google Messages
Interfaces
desk · Desktop GUI mob · Mobile touch
GUI need · task setting
G3 · GUI-only
GUI need · ordinary access
G3 · GUI-only
Interaction primitives
click · Point & click type · Text & forms multi · Cross-app
Work product
WP08 · App/system state
Conditions
xdevice · Cross-device risk · Irreversible/risky
Proposed difficulty
4 / 5

Proposed verifier

Emulator mmssms.db outgoing to Lisa with exact body; calllog outgoing to the signature number. Emulator input audit shows no touch events during the run. Phone Link has no API.

Proposed partial credit

40% exact SMS reply; 40% call to the signature number; 20% no phone-side input

R11.01.01.05-T12 Star and batch-forward flight messages from a WhatsApp family group

Original instruction

WhatsApp Desktop (test number) shows group 'Family' with 2,000 messages. Star every October 2026 message that contains a flight number (two letters plus 3-4 digits), then forward all of them to 'Dad' in one batch in chronological order. Ignore media without text.

Platforms
win · Windows mac · macOS
Applications
WhatsApp Desktop
Interfaces
desk · Desktop GUI
GUI need · task setting
G3 · GUI-only
GUI need · ordinary access
G3 · GUI-only
Interaction primitives
nav · Pan/zoom/scroll click · Point & click key · Shortcuts
Work product
WP08 · App/system state
Conditions
long · Long-horizon risk · Irreversible/risky
Proposed difficulty
4 / 5

Proposed verifier

Linked rooted phone msgstore.db: starred set vs expected (F1); Dad's emulator receives forwarded-flag messages with the right bodies in order. WhatsApp has no consumer API, and the agent holds no linked-device session other than the desktop client, so stars and forwards come only from it.

Proposed partial credit

50% starred F1; 30% forwarded set matches; 20% order preserved

R11.01.01.05-T13 Tapback, reply in thread and edit a sent iMessage

Original instruction

In macOS Messages (test Apple ID), chat with Jordan: heart the three messages that contain a date, reply in thread to the question 'Which gate?' with 'B32', then edit your most recent sent message (sent by the harness at task start) to change 'tomorow' to 'tomorrow'.

Platforms
mac · macOS
Applications
Messages
Interfaces
desk · Desktop GUI
GUI need · task setting
G3 · GUI-only
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
hover · Hover & context click · Point & click type · Text & forms
Work product
WP08 · App/system state
Conditions
risk · Irreversible/risky
Proposed difficulty
3 / 5

Proposed verifier

chat.db: exactly three love reactions targeting those message guids; reply with thread_originator_guid of the question and text 'B32'; edited message has the corrected text and an edit record. No AppleScript route exists for these, and Terminal lacks Full Disk Access, so chat.db cannot be edited from a shell.

Proposed partial credit

40% correct three hearts and no extras; 30% threaded reply; 30% edit applied

R11.01.01.05-T14 Block and report scam SMS senders but keep the real courier thread

Original instruction

On the Android emulator, 5 recent threads mention a 'parcel fee'. Three are scams (shortened links, sender not the courier short code in fixtures/couriers.txt). Block and report those three as spam and delete their threads; keep the legitimate courier thread and do not open any link.

Platforms
and · Android
Applications
Google Messages
Interfaces
mob · Mobile touch
GUI need · task setting
G2 · GUI-essential
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
touch · Touch gestures vis · Visual judgment
Work product
WP08 · App/system state
Conditions
risk · Irreversible/risky
Proposed difficulty
3 / 5

Proposed verifier

BlockedNumberProvider holds exactly the 3 scam senders; their threads are gone from mmssms.db; courier thread intact; proxy log shows no request to the scam domains. adb is not exposed to the agent, so actions come only from emulator UI input.

Proposed partial credit

45% correct blocks (minus 15 points per wrong block); 25% scam threads deleted; 20% courier thread kept; 10% no link opened

R11.01.01.05-T15 Control pair of T01: send the SMS reply from a shell over adb

Original instruction

Control pair of T01. With only a terminal and adb connected to the same emulator state, send 'On my way' to +1-555-0100 without the GUI (an intent with auto-send, or the telephony service) and confirm it is stored as sent.

Platforms
lin · Linux and · Android
Applications
adb
Interfaces
term · Terminal
GUI need · task setting
CL · CLI-sufficient
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
code · Code & commands
Work product
WP08 · App/system state
Conditions
risk · Irreversible/risky
Proposed difficulty
2 / 5

Proposed verifier

Same mmssms.db check as T01; input audit shows no touch events. Result compared with T01 to measure the GUI/CLI gap.

Proposed partial credit

70% correct outgoing SMS stored as sent; 30% no GUI input used

R11.01.01.05-T16 Send one group MMS with a cropped receipt photo to three friends

Original instruction

On the Android emulator, start one group conversation (not three separate texts) with Ana, Ben and Chloe. Attach the newest photo in Google Photos cropped to just the restaurant receipt with the editor, add the text 'Split 3 ways: $27.40 each', and send it as group MMS.

Platforms
and · Android
Applications
Google Messages · Google Photos
Interfaces
mob · Mobile touch
GUI need · task setting
G2 · GUI-essential
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
touch · Touch gestures drag · Drag & drop vis · Visual judgment
Work product
WP08 · App/system state
Conditions
multiapp · Multi-app risk · Irreversible/risky
Proposed difficulty
3 / 5

Proposed verifier

mmssms.db: one MMS thread with 3 recipients; its image part, template-matched to the original, covers the receipt with IoU >= 0.8; text exact. Three separate SMS score 0 on milestone 1. adb is not exposed to the agent, so actions come only from emulator UI input.

Proposed partial credit

35% single group thread with 3 recipients; 35% cropped receipt (IoU); 30% exact text

R11.01.01.05-T17 Forward chosen WeChat messages as one merged chat record

Original instruction

In WeChat for Windows (benchmark test account), select the 8 messages in group 'Renovation' from 2026-09-30 about the tile order, skipping stickers and unrelated replies, and forward them to contact 'Lin' as one merged chat record (not one by one), then send the note 'Please check quantity'.

Platforms
win · Windows mac · macOS
Applications
WeChat
Interfaces
desk · Desktop GUI
GUI need · task setting
G3 · GUI-only
GUI need · ordinary access
G3 · GUI-only
Interaction primitives
click · Point & click nav · Pan/zoom/scroll vis · Visual judgment
Work product
WP08 · App/system state
Conditions
risk · Irreversible/risky
Proposed difficulty
4 / 5

Proposed verifier

Recipient test phone DB (harness-decrypted): one merged chat-record message containing exactly the 8 source messages in order, plus the text note. One-by-one forwards get partial credit only. No API exists.

Proposed partial credit

40% merged-record format; 40% exactly the 8 messages (F1); 20% note sent

R11.01.01.05-T18 Coordinate a dinner from a group chat and confirm with a map link

Original instruction

On the Android emulator with a WhatsApp test number, three test contacts replied in WhatsApp group 'Fri Dinner' with free times and dietary limits. Find the only 1-hour Friday slot all can make, pick the restaurant from fixtures/shortlist.csv that fits every limit and is open then, add a Calendar event, and post one confirmation with time, restaurant and its Maps link.

Platforms
and · Android
Applications
WhatsApp · Google Maps · Google Calendar
Interfaces
mob · Mobile touch
GUI need · task setting
G3 · GUI-only
GUI need · ordinary access
G3 · GUI-only
Interaction primitives
touch · Touch gestures multi · Cross-app type · Text & forms
Work product
WP08 · App/system state
Conditions
multiapp · Multi-app risk · Irreversible/risky
Proposed difficulty
5 / 5

Proposed verifier

Group receives one message with the correct slot, restaurant name and a maps URL resolving to its place ID; Calendar event matches slot and location. Wrong slot or restaurant zeroes that milestone. WhatsApp has no consumer API and adb is not exposed to the agent, so the message is posted only from the emulator UI.

Proposed partial credit

30% correct slot; 30% correct restaurant; 20% calendar event; 20% confirmation message with valid map link

R11.01.01.05-T19 Return family missed calls and save an unknown caller as a contact

Original instruction

On the Android emulator, Phone shows 6 missed calls today. Call back only the numbers in Contacts group 'Family' (hang up once connected), and save the unknown number as contact 'Plumber - Ortiz' with the company and email from the SMS that number sent at 10:41.

Platforms
and · Android
Applications
Phone · Contacts · Google Messages
Interfaces
mob · Mobile touch
GUI need · task setting
G2 · GUI-essential
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
touch · Touch gestures type · Text & forms multi · Cross-app
Work product
WP08 · App/system state
Conditions
multiapp · Multi-app risk · Irreversible/risky
Proposed difficulty
3 / 5

Proposed verifier

calllog.db: outgoing calls to exactly the Family-group missed numbers; contacts2.db has a new contact with name, number, company and email from that SMS; no calls to anyone else. adb is not exposed to the agent, so actions come only from emulator UI input.

Proposed partial credit

45% correct callbacks (minus 15 points per wrong call); 40% new contact fields; 15% no extra calls

R11.01.01.05-T20 Collect contractor invoices from iMessage and reply with the total

Original instruction

In macOS Messages (test Apple ID), the chat with 'Contractor Lee' spans 9 months. Save every invoice PDF he sent since 2026-06-01 to ~/Documents/Renovation/Invoices as YYYY-MM-DD_<invoice no>.pdf, skipping quotes and photos, then reply to him with the sum of the invoice totals, e.g. 'Total received: $12,345.67'.

Platforms
mac · macOS
Applications
Messages · Finder · Preview
Interfaces
desk · Desktop GUI
GUI need · task setting
G2 · GUI-essential
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
nav · Pan/zoom/scroll drag · Drag & drop multi · Cross-app
Work product
WP09 · Files & folders
Conditions
long · Long-horizon risk · Irreversible/risky
Proposed difficulty
5 / 5

Proposed verifier

Folder holds exactly the expected files (SHA-256 equal to originals, exact names); Lee's chat has an outgoing message with the correct total to the cent. Terminal lacks Full Disk Access, so ~/Library/Messages is unreadable from a shell; eslogger detects any such read (score 0).

Proposed partial credit

40% files saved (F1); 20% filenames; 30% correct total in reply; 10% no quotes or photos saved

Mapped research references

Existing benchmark-to-class links from the research catalog. They are not record-level annotations, executions or product endorsements.

In this workflow family