Computer-Use Agent Taxonomy中文
R05.05.01.03 · TASK CLASS

Legacy terminal & virtual desktop entry

Enter and verify data through 3270/5250 emulators and Citrix/RDP sessions.

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

Typical applications
IBM terminal emulators · Citrix Workspace · Remote Desktop
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.

R05.05.01.03-T01 Look up a policy's mailing address on a 3270 screen

Original instruction

IBM Personal Communications (Windows) session to the sandbox mainframe POLICY application (CICS-style). Log in with input/creds_test.txt, use transaction PINQ to display policy P-00412, and write its mailing address lines and ZIP to output/address.txt exactly as shown.

Platforms
win · Windows
Applications
IBM Personal Communications
Interfaces
desk · Desktop GUI
GUI need · task setting
G3 · GUI-only
GUI need · ordinary access
G3 · GUI-only
Interaction primitives
type · Text & forms key · Shortcuts
Work product
WP12 · Answer/decision
Conditions
licensed · Licensed pro app
Proposed difficulty
1 / 5

Proposed verifier

Text equals the host DB record (trailing spaces trimmed). Host transaction log must show PINQ for P-00412 from the agent's LU. TN3270/TN5250 ports are firewalled from the agent shell and emulator macros/HLLAPI are disabled, so only the GUI emulator reaches the host.

Proposed partial credit

70% address lines; 30% ZIP

R05.05.01.03-T02 Change a customer's credit limit on an IBM i 5250 screen

Original instruction

IBM i Access Client Solutions 5250 emulator (Java) to the sandbox order-entry app. Using option 2 on the customer subfile (CUSTMNT), change customer 004217's credit limit to 25,000 and credit-hold flag to N, then press F10 to save; confirm the message line.

Platforms
win · Windows mac · macOS lin · Linux
Applications
IBM i Access Client Solutions
Interfaces
desk · Desktop GUI
GUI need · task setting
G3 · GUI-only
GUI need · ordinary access
G3 · GUI-only
Interaction primitives
type · Text & forms key · Shortcuts
Work product
WP08 · App/system state
Conditions
None declared
Proposed difficulty
2 / 5

Proposed verifier

DB2 for i table CUSTMAST: CRLIMIT 25000, HOLD 'N' for 004217; journal entry by agent's user profile via interactive job; no other rows changed. TN3270/TN5250 ports are firewalled from the agent shell and emulator macros/HLLAPI are disabled, so only the GUI emulator reaches the host.

Proposed partial credit

50% credit limit; 30% hold flag; 20% no collateral changes

R05.05.01.03-T03 Enter one stock adjustment in a legacy app over Remote Desktop

Original instruction

Microsoft Remote Desktop / Windows App to a Windows Server 2012 R2 host running the VB6 app 'InvTrack'; clipboard and drive redirection are disabled. Enter a stock adjustment of -14 units for SKU BRK-200 in location A-12, reason 'Damaged', reference 'ADJ-1014', and save.

Platforms
mac · macOS win · Windows
Applications
Remote Desktop · InvTrack (VB6)
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 vis · Visual judgment
Work product
WP08 · App/system state
Conditions
None declared
Proposed difficulty
2 / 5

Proposed verifier

InvTrack SQL Server table adjustments: one new row with values as specified, created by the RDP session user; on-hand decreased by 14. RDP clipboard/drive redirection disabled server-side.

Proposed partial credit

60% adjustment row correct; 20% reason/reference; 20% on-hand updated

R05.05.01.03-T04 Key fifteen policy address changes into a 3270 update screen

Original instruction

wc3270 (Windows) session to the sandbox POLICY app. input/changes.xlsx lists 15 policy address changes (some with 2 address lines, some with apartment numbers). For each, use transaction PUPD, tab through protected fields, enter the new address, effective date 2026-10-15, and press PF5 to commit; verify the 'UPDATE COMPLETE' message.

Platforms
win · Windows
Applications
wc3270 · Excel
Interfaces
desk · Desktop GUI
GUI need · task setting
G3 · GUI-only
GUI need · ordinary access
G3 · GUI-only
Interaction primitives
type · Text & forms key · Shortcuts multi · Cross-app
Work product
WP08 · App/system state
Conditions
multiapp · Multi-app
Proposed difficulty
3 / 5

Proposed verifier

Host VSAM/DB rows for the 15 policies equal key; effective date set; host log shows PUPD commits from the agent's terminal session. TN3270/TN5250 ports are firewalled from the agent shell and emulator macros/HLLAPI are disabled, so only the GUI emulator reaches the host. Control pair: T12.

Proposed partial credit

70% per-policy address exact; 20% effective dates; 10% no other policies changed

R05.05.01.03-T05 Post eight invoices through a Citrix-published accounting app

Original instruction

Citrix Workspace app launching the published application 'LedgerPro 7' (Windows legacy accounting). input/invoices_oct.csv has 8 vendor invoices. Enter each in AP > Invoice Entry with vendor code lookup (F2), GL distribution and due date, then post the batch. Record the batch number in output/batch.txt.

Platforms
mac · macOS win · Windows
Applications
Citrix Workspace · LedgerPro (published app)
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 key · Shortcuts vis · Visual judgment
Work product
WP08 · App/system state
Conditions
licensed · Licensed pro app
Proposed difficulty
3 / 5

Proposed verifier

LedgerPro DB on the VDA: batch with 8 posted invoices equal CSV (vendor, amount, GL, due date); batch number matches. Citrix clipboard, drive mapping and file transfer are disabled; the inner app is pixels only (no accessibility tree).

Proposed partial credit

60% invoices exact (per invoice); 25% batch posted; 15% batch number

R05.05.01.03-T06 Transcribe twenty account balances from terminal screens into Excel

Original instruction

x3270 (Linux) or c3270 (macOS) to the sandbox BANKINQ app. For the 20 customer IDs in input/balances.xlsx, display each account summary (transaction BINQ, PF8 for the second page when the account has multiple products) and fill columns C-E with available balance, ledger balance and hold amount. Save the workbook.

Platforms
lin · Linux mac · macOS
Applications
x3270 · Excel
Interfaces
desk · Desktop GUI term · Terminal
GUI need · task setting
G3 · GUI-only
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
type · Text & forms key · Shortcuts multi · Cross-app
Work product
WP02 · Records/table
Conditions
multiapp · Multi-app
Proposed difficulty
3 / 5

Proposed verifier

60 cells equal host values at snapshot time; multi-product accounts aggregated as specified on the second page. TN3270/TN5250 ports are firewalled from the agent shell and emulator macros/HLLAPI are disabled, so only the GUI emulator reaches the host; screen-scraping scripts (s3270) are not installed.

Proposed partial credit

70% balances exact (per cell); 30% multi-page accounts correct

R05.05.01.03-T07 Page through a 5250 subfile to prioritize old backorders

Original instruction

Mocha TN5250 (Windows) or IBM i ACS to the sandbox order app. In ORDINQ, filter status 'BO', then roll through the subfile (Page Down) to find all backorder lines older than 30 days as of 2026-10-14. Use option 2 to set ship priority 'A' on each and F10 to save.

Platforms
win · Windows mac · macOS
Applications
Mocha TN5250 · IBM i Access Client Solutions
Interfaces
desk · Desktop GUI
GUI need · task setting
G3 · GUI-only
GUI need · ordinary access
G3 · GUI-only
Interaction primitives
type · Text & forms key · Shortcuts nav · Pan/zoom/scroll
Work product
WP08 · App/system state
Conditions
None declared
Proposed difficulty
3 / 5

Proposed verifier

ORDDTL rows: priority 'A' on exactly the key's lines (across 6 subfile pages); others unchanged; interactive job journal by agent. TN3270/TN5250 ports are firewalled from the agent shell and emulator macros/HLLAPI are disabled, so only the GUI emulator reaches the host.

Proposed partial credit

60% exact set prioritized; 25% none missed on later pages; 15% no collateral changes

R05.05.01.03-T08 Terminate an employee record in a legacy HR app over VMware Horizon

Original instruction

VMware Horizon client to a Windows 10 desktop running the Delphi HR app 'PersonnelPlus'. Per input/termination_form.pdf, terminate employee 10842 effective 2026-10-31, reason code VOL-RES, final pay date, and set rehire-eligible Y. Stop at the 'Process final payroll?' prompt and answer No.

Platforms
mac · macOS win · Windows
Applications
VMware Horizon · PersonnelPlus
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 vis · Visual judgment
Work product
WP08 · App/system state
Conditions
risk · Irreversible/risky
Proposed difficulty
3 / 5

Proposed verifier

HR DB: employee 10842 status T with effective date, reason, rehire flag equal form; no final-payroll job queued. Horizon clipboard and USB redirection disabled; app has no API.

Proposed partial credit

50% termination fields; 20% rehire flag; 30% no payroll job queued

R05.05.01.03-T09 Run a parts lookup and reorder in an Access front-end inside Azure Virtual Desktop

Original instruction

Azure Virtual Desktop (Windows App client) session with the Access 2010 runtime app 'PartsDesk'. For the 6 parts in input/reorder.txt, open each in the Parts form, read on-hand and reorder point, and create a reorder for any part below reorder point using the 'Reorder' button with quantity = max level - on hand.

Platforms
mac · macOS win · Windows
Applications
Azure Virtual Desktop · Microsoft Access
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 vis · Visual judgment
Work product
WP08 · App/system state
Conditions
None declared
Proposed difficulty
3 / 5

Proposed verifier

Back-end .accdb on the session host: reorder rows exist for exactly the key parts with quantities equal key; created by the AVD session user. The RemoteApp publishes only PartsDesk (no desktop, cmd or PowerShell) and the back-end is not reachable from the agent's machine (no share or ODBC route); clipboard redirection is off and the runtime hides the navigation pane and VBA editor.

Proposed partial credit

50% correct parts reordered; 40% quantities; 10% none for parts above reorder point

R05.05.01.03-T10 Adjudicate five claims on a CICS-style 3270 screen

Original instruction

IBM Personal Communications to the sandbox CLAIMS app. Adjudicate the 5 pended claims in input/claims_queue.txt with transaction CLMA: apply the rules in input/adjudication_rules.pdf (deductible, coinsurance, out-of-network reduction), enter allowed amounts and decision codes, and PF6 to finalize each.

Platforms
win · Windows
Applications
IBM Personal Communications
Interfaces
desk · Desktop GUI
GUI need · task setting
G3 · GUI-only
GUI need · ordinary access
G3 · GUI-only
Interaction primitives
type · Text & forms key · Shortcuts
Work product
WP08 · App/system state
Conditions
licensed · Licensed pro app
Proposed difficulty
3 / 5

Proposed verifier

Host CLAIMS records: decision codes and allowed/paid amounts equal key per claim; finalize timestamps from the agent's LU. TN3270/TN5250 ports are firewalled from the agent shell and emulator macros/HLLAPI are disabled, so only the GUI emulator reaches the host.

Proposed partial credit

60% amounts per claim; 30% decision codes; 10% finalized

R05.05.01.03-T11 Enter twelve shipments in a VT220 warehouse app through a flaky RDP link

Original instruction

Remote Desktop to a jump host whose PuTTY session runs the VT220 WMS 'ShipRite'. The RDP link is configured to drop twice during the task. Enter the 12 shipments in input/shipments.csv (screen SHP-01, function keys F2 save, F7 next), reconnecting as needed, and make sure none is lost or duplicated.

Platforms
mac · macOS win · Windows
Applications
Remote Desktop · PuTTY
Interfaces
desk · Desktop GUI
GUI need · task setting
G3 · GUI-only
GUI need · ordinary access
G3 · GUI-only
Interaction primitives
type · Text & forms key · Shortcuts
Work product
WP08 · App/system state
Conditions
interrupt · Interruptions
Proposed difficulty
3 / 5

Proposed verifier

WMS DB: exactly 12 shipments equal CSV (no duplicates); RDP reconnect events present in host logs. The WMS host's SSH port is firewalled except from the RDP server, where clipboard/drive redirection and PuTTY session logging are disabled, so shipments must be typed in the PuTTY window.

Proposed partial credit

70% shipments exact; 20% no duplicates after reconnects; 10% all saved

R05.05.01.03-T12 Script the same address changes with s3270 (control pair of T04)

Original instruction

Control pair of T04. On a Linux host with s3270 and py3270 allowed to reach the same sandbox (env TN_HOST), script the 15 PUPD address changes from input/changes.xlsx, verifying each 'UPDATE COMPLETE' message.

Platforms
lin · Linux
Applications
s3270 · py3270
Interfaces
term · Terminal
GUI need · task setting
CL · CLI-sufficient
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
code · Code & commands
Work product
WP08 · App/system state
Conditions
None declared
Proposed difficulty
3 / 5

Proposed verifier

Same host-row checks as T04; used to measure the GUI penalty on identical content.

Proposed partial credit

70% per-policy address; 20% dates; 10% no other changes

R05.05.01.03-T13 Process six policy endorsements with conditional screens and edit errors

Original instruction

wc3270 to the sandbox POLICY app. Apply the 6 endorsements in input/endorsements.pdf (vehicle add, driver removal, coverage limit change). Each flows through 2-4 screens depending on type, with host edits (e.g. 'VIN CHECK DIGIT INVALID' for one transposed VIN you must correct from the vehicle title image). Commit each with PF5.

Platforms
win · Windows
Applications
wc3270 · Adobe Acrobat Reader
Interfaces
desk · Desktop GUI
GUI need · task setting
G3 · GUI-only
GUI need · ordinary access
G3 · GUI-only
Interaction primitives
type · Text & forms key · Shortcuts vis · Visual judgment
Work product
WP08 · App/system state
Conditions
None declared
Proposed difficulty
4 / 5

Proposed verifier

Host policy records reflect all 6 endorsements per key, including the corrected VIN; premium recalculation fields equal key. TN3270/TN5250 ports are firewalled from the agent shell and emulator macros/HLLAPI are disabled, so only the GUI emulator reaches the host.

Proposed partial credit

60% six endorsements applied (equal shares); 20% VIN fix; 20% premiums

R05.05.01.03-T14 Copy month-end totals from a double-hop legacy server into local Excel

Original instruction

Citrix Workspace desktop 'FIN-VDI' that itself opens an RDP session to legacy server LEG-02 running the 'GLReports' app. Run the Month-End Trial Balance for September 2026, then transcribe the 24 account totals into local input/tb_capture.xlsx (no clipboard or file transfer at either hop).

Platforms
mac · macOS win · Windows
Applications
Citrix Workspace · Remote Desktop · Excel
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 vis · Visual judgment multi · Cross-app
Work product
WP02 · Records/table
Conditions
licensed · Licensed pro app multiapp · Multi-app
Proposed difficulty
4 / 5

Proposed verifier

24 cells equal report values from the legacy DB; report run logged on LEG-02 by the session user. Citrix clipboard, drive mapping and file transfer are disabled; the inner app is pixels only (no accessibility tree); inner RDP has redirection disabled too.

Proposed partial credit

70% totals exact (per account); 20% debits = credits check passes; 10% report run

R05.05.01.03-T15 Update forty item prices on an IBM i maintenance screen from a PDF list

Original instruction

IBM i ACS 5250 to the sandbox pricing app (PRCMNT). input/price_list_nov.pdf is a supplier price list; update the 40 items' list price and effective date 2026-11-01, using F4 prompts for unit of measure where the PDF uses a different UoM (convert per input/uom.xlsx). Save each with F10.

Platforms
win · Windows mac · macOS lin · Linux
Applications
IBM i Access Client Solutions · Adobe Acrobat Reader
Interfaces
desk · Desktop GUI
GUI need · task setting
G3 · GUI-only
GUI need · ordinary access
G3 · GUI-only
Interaction primitives
type · Text & forms key · Shortcuts multi · Cross-app
Work product
WP08 · App/system state
Conditions
long · Long-horizon
Proposed difficulty
4 / 5

Proposed verifier

PRICEF rows: 40 prices equal key after UoM conversion, effective dates set; no other items changed; journal by interactive job. TN3270/TN5250 ports are firewalled from the agent shell and emulator macros/HLLAPI are disabled, so only the GUI emulator reaches the host.

Proposed partial credit

60% prices (per item); 25% UoM conversions; 15% dates and no collateral

R05.05.01.03-T16 Enter journal lines with LOVs in a Citrix-published Oracle Forms app

Original instruction

Citrix Workspace launching published Oracle E-Business Suite 11i Forms (Java). In GL > Journals > Enter, create journal 'SEP-ACCR-07' in the Sept-26 period with 6 lines from input/je_lines.xlsx, using List of Values for each account segment combination, then reserve funds and save (do not post).

Platforms
mac · macOS win · Windows
Applications
Citrix Workspace · Oracle E-Business Suite
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 key · Shortcuts vis · Visual judgment
Work product
WP08 · App/system state
Conditions
licensed · Licensed pro app
Proposed difficulty
4 / 5

Proposed verifier

GL_JE_HEADERS/LINES: journal with 6 lines equal key code combinations and amounts; status unposted; created by the agent's EBS user. Citrix clipboard, drive mapping and file transfer are disabled; the inner app is pixels only (no accessibility tree).

Proposed partial credit

60% lines exact; 20% header/period; 20% saved not posted

R05.05.01.03-T17 Enter twenty records across RDP idle timeouts without losing work

Original instruction

Remote Desktop to a host whose legacy app 'FieldLog' discards unsaved forms when the 5-minute idle timeout disconnects the session (a scheduled disconnect occurs at least twice). Enter the 20 inspection records in input/inspections.csv, saving in small batches, and after each reconnect check which records exist before continuing.

Platforms
mac · macOS win · Windows
Applications
Remote Desktop · FieldLog
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 vis · Visual judgment
Work product
WP08 · App/system state
Conditions
interrupt · Interruptions hidden · Hidden/stale state
Proposed difficulty
4 / 5

Proposed verifier

FieldLog DB: exactly 20 records equal CSV; no duplicates; forced-disconnect events in logs prove the interruptions happened. RDP clipboard and drive redirection are disabled server-side and the FieldLog database is firewalled from the agent machine, so records must be keyed in the remote app.

Proposed partial credit

70% records exact; 20% no duplicates; 10% completed after interruptions

R05.05.01.03-T18 Key twenty-five paper claim forms into the 3270 claims system

Original instruction

IBM Personal Communications to the sandbox CLAIMS app. input/claim_forms/ has 25 scanned CMS-1500-style forms (some handwritten). Key each with transaction CLME (header, up to 6 service lines, diagnosis codes), resolve host edit messages, and pend claims with missing data using pend code M07 rather than guessing.

Platforms
win · Windows
Applications
IBM Personal Communications · Adobe Acrobat Reader
Interfaces
desk · Desktop GUI
GUI need · task setting
G3 · GUI-only
GUI need · ordinary access
G3 · GUI-only
Interaction primitives
type · Text & forms key · Shortcuts vis · Visual judgment
Work product
WP08 · App/system state
Conditions
licensed · Licensed pro app long · Long-horizon
Proposed difficulty
5 / 5

Proposed verifier

Host claim records: per-claim fields equal key (>= 98% field accuracy for full credit); the key's 3 incomplete forms pended with M07; no invented values. TN3270/TN5250 ports are firewalled from the agent shell and emulator macros/HLLAPI are disabled, so only the GUI emulator reaches the host.

Proposed partial credit

60% field accuracy; 20% correct pends; 20% service-line totals match

R05.05.01.03-T19 Reconcile AS/400 inventory inside Citrix against a warehouse count

Original instruction

Citrix desktop 'OPS-VDI' with IBM i ACS 5250 to the inventory app. Compare the 60 SKUs on input/count_sheet.pdf (local) to on-hand in screen INVINQ, and for differences over 2 units post an adjustment in INVADJ with reason CC (cycle count); write output/variance.csv locally with SKU, system, counted, posted.

Platforms
mac · macOS win · Windows
Applications
Citrix Workspace · IBM i Access Client Solutions · Excel
Interfaces
desk · Desktop GUI
GUI need · task setting
G3 · GUI-only
GUI need · ordinary access
G3 · GUI-only
Interaction primitives
type · Text & forms key · Shortcuts vis · Visual judgment multi · Cross-app
Work product
WP08 · App/system state
Conditions
licensed · Licensed pro app long · Long-horizon multiapp · Multi-app
Proposed difficulty
5 / 5

Proposed verifier

Host adjustments equal key set and quantities (only |diff| > 2); variance.csv rows equal key. Citrix clipboard, drive mapping and file transfer are disabled; the inner app is pixels only (no accessibility tree); TN3270/TN5250 ports are firewalled from the agent shell and emulator macros/HLLAPI are disabled, so only the GUI emulator reaches the host.

Proposed partial credit

50% adjustments correct; 20% threshold respected; 30% variance file

R05.05.01.03-T20 Migrate thirty customer records from a 3270 app to a web CRM by hand

Original instruction

Remote Desktop to jump host JH-1, which runs wc3270 to the sandbox CUSTMAST app. For the 30 customer IDs in input/migrate.txt, read name, address, phone, credit terms and 3 recent order totals (two screens each), and create matching accounts in the local browser's sandbox CRM (no import tool), mapping terms codes via input/terms_map.csv.

Platforms
mac · macOS win · Windows
Applications
Remote Desktop · wc3270 · Sandbox CRM
Interfaces
desk · Desktop GUI brow · Browser
GUI need · task setting
G3 · GUI-only
GUI need · ordinary access
G3 · GUI-only
Interaction primitives
type · Text & forms key · Shortcuts multi · Cross-app
Work product
WP08 · App/system state
Conditions
multiapp · Multi-app long · Long-horizon
Proposed difficulty
5 / 5

Proposed verifier

CRM DB: 30 accounts with fields equal host data after mapping (>= 98% field accuracy for full credit); CRM import endpoint disabled; RDP clipboard off. TN3270/TN5250 ports are firewalled from the agent shell and emulator macros/HLLAPI are disabled, so only the GUI emulator reaches the host.

Proposed partial credit

60% field accuracy; 20% term mapping; 20% order totals

Mapped research references

No benchmark reference is assigned to this class in the current mapping.

In this workflow family