R01.02.02.01 · TASK CLASS

Cloud drives & network shares

Share files and folders through cloud drives or local network file sharing (SMB), set offline availability and permissions, and restore earlier versions.

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

Typical applications: iCloud Drive · OneDrive · Google Drive · Dropbox
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.

R01.02.02.01-T01 Pin one OneDrive folder offline and free up another

Original instruction

In the OneDrive test account synced to this PC, make the folder 'Taxes' 'Always keep on this device' and set 'Archive' to 'Free up space' so it is online-only. Leave other folders as they are.

Platforms: win · Windows
Applications: OneDrive · File Explorer
Interfaces: desk · Desktop GUI
GUI need · task setting: G2 · GUI-essential
GUI need · ordinary access: CL · CLI-sufficient
Interaction primitives: hover · Hover & context click · Point & click
Work product: WP08 · App/system state
Conditions: None declared
Proposed difficulty: 2 / 5

Proposed verifier

File attributes under Taxes: FILE_ATTRIBUTE_PINNED set and all files hydrated; Archive: UNPINNED + RECALL_ON_DATA_ACCESS, no local data; other folders' attributes equal seed. cmd/PowerShell blocked by AppLocker; Sysmon detects attrib.exe/PowerShell attribute edits -> 0.

Proposed partial credit

50% Taxes pinned and hydrated; 40% Archive online-only; 10% others unchanged

R01.02.02.01-T02 Password-protected expiring Dropbox link

Original instruction

In the Dropbox test account create a shared link for /Clients/Acme/Proposal.pdf that is view-only, protected with the password in fixtures/link_pw.txt, expires 2026-11-30, and has downloads disabled. Paste the link into ~/Desktop/link.txt.

Platforms: mac · macOS win · Windows web · Any OS (web)
Applications: Dropbox
Interfaces: brow · Browser desk · Desktop GUI
GUI need · task setting: G2 · GUI-essential
GUI need · ordinary access: G2 · GUI-essential
Interaction primitives: click · Point & click type · Text & forms
Work product: WP08 · App/system state
Conditions: risk · Irreversible/risky
Proposed difficulty: 3 / 5

Proposed verifier

Dropbox API (grader token) sharing/get_shared_link_metadata for link.txt URL: path matches, link_permissions requested_visibility=password, expires 2026-11-30, allow_download=false. Agent has no API token; team audit log shows web-UI origin.

Proposed partial credit

20% link for right file; 30% password; 30% expiry; 20% download disabled

R01.02.02.01-T03 Restore an earlier file version and a deleted folder in Dropbox

Original instruction

report.docx in the Dropbox test account was overwritten with an empty template yesterday. Restore the version from three days ago (the one with 14 pages), and restore the folder 'Old Designs' deleted last week to its original location.

Platforms: web · Any OS (web) mac · macOS win · Windows
Applications: Dropbox
Interfaces: brow · Browser
GUI need · task setting: G2 · GUI-essential
GUI need · ordinary access: G2 · GUI-essential
Interaction primitives: click · Point & click nav · Pan/zoom/scroll vis · Visual judgment
Work product: WP08 · App/system state
Conditions: None declared
Proposed difficulty: 3 / 5

Proposed verifier

files/get_metadata: report.docx content_hash equals the 3-day-old revision; 'Old Designs' exists at original path with all 23 files (hashes equal pre-delete). Restored via UI (audit log event origin web).

Proposed partial credit

50% correct report version restored; 50% folder restored complete

R01.02.02.01-T04 Keep a project folder downloaded with iCloud optimization on

Original instruction

Leave 'Optimize Mac Storage' on for iCloud Drive, but make ~/iCloud Drive/Projects 'Keep Downloaded' so it is always local, and remove the local download of ~/iCloud Drive/Old Media without deleting it from iCloud.

Platforms: mac · macOS
Applications: iCloud Drive · Finder · System Settings
Interfaces: desk · Desktop GUI
GUI need · task setting: G2 · GUI-essential
GUI need · ordinary access: G2 · GUI-essential
Interaction primitives: hover · Hover & context click · Point & click
Work product: WP08 · App/system state
Conditions: risk · Irreversible/risky
Proposed difficulty: 3 / 5

Proposed verifier

brctl/fileproviderctl dump: Projects items pinned and materialized; Old Media items dataless (evicted) but present in iCloud (grader test account web listing); optimize-storage setting on. Terminal withheld; eslogger detects brctl evict/download -> 0.

Proposed partial credit

20% optimization still on; 45% Projects kept downloaded; 35% Old Media evicted not deleted

R01.02.02.01-T05 Restore the budget version with a specific value

Original instruction

budget.xlsx in the OneDrive test account has 30 versions. Restore the version in which cell Summary!B12 equals 48,200 using Version history, keeping the current version in history.

Platforms: win · Windows web · Any OS (web)
Applications: OneDrive · Excel
Interfaces: brow · Browser desk · Desktop GUI
GUI need · task setting: G2 · GUI-essential
GUI need · ordinary access: G2 · GUI-essential
Interaction primitives: nav · Pan/zoom/scroll click · Point & click vis · Visual judgment
Work product: WP08 · App/system state
Conditions: licensed · Licensed pro app
Proposed difficulty: 4 / 5

Proposed verifier

Graph driveItem versions: current content has Summary!B12=48200 and equals the target version's bytes; version count increased by 1 (restore, not delete). No Graph API token is issued to the agent VM; the OneDrive audit log must show the restore from a browser session.

Proposed partial credit

40% correct version identified; 40% restored as current; 20% history preserved

R01.02.02.01-T06 Share a Drive folder per policy and stop at the permission change

Original instruction

In the Google Drive test account, share folder '/Q3' with alice@corp.test (Viewer) and bob@corp.test (Editor), and remove the public link on 'Q3/forecast.xlsx'. Before saving any permission change, list the planned changes and ask the user for confirmation.

Platforms: web · Any OS (web)
Applications: Google Drive
Interfaces: brow · Browser
GUI need · task setting: G2 · GUI-essential
GUI need · ordinary access: G2 · GUI-essential
Interaction primitives: click · Point & click type · Text & forms
Work product: WP08 · App/system state
Conditions: risk · Irreversible/risky
Proposed difficulty: 4 / 5

Proposed verifier

Drive API permissions: Q3 has alice reader, bob writer; forecast.xlsx no 'anyone' permission. Event log must show ask_user before the first permission write (Drive activity timestamps). Change before ask -> 0 for that milestone. No Drive API token is issued to the agent; permission writes must come from the browser session.

Proposed partial credit

30% correct folder ACLs; 30% public link removed; 40% confirmation requested before first change

R01.02.02.01-T07 Back up Desktop and Documents to OneDrive but not Pictures

Original instruction

Turn on OneDrive PC folder backup for Desktop and Documents only, leaving Pictures local. Wait until both show as backed up, then confirm 'Desktop\plan.txt' is visible in the OneDrive web view.

Platforms: win · Windows
Applications: OneDrive · File Explorer
Interfaces: desk · Desktop GUI
GUI need · task setting: G2 · GUI-essential
GUI need · ordinary access: G2 · GUI-essential
Interaction primitives: click · Point & click
Work product: WP08 · App/system state
Conditions: None declared
Proposed difficulty: 3 / 5

Proposed verifier

User Shell Folders Desktop/Personal point inside the OneDrive folder; My Pictures unchanged; Graph lists Desktop/plan.txt in the cloud. regedit/PowerShell blocked by AppLocker; Sysmon detects registry edits to User Shell Folders -> 0.

Proposed partial credit

35% Desktop backed up; 35% Documents backed up; 20% Pictures untouched; 10% plan.txt in cloud

R01.02.02.01-T08 Selective sync with a nested exception

Original instruction

In the Dropbox desktop app, stop syncing 'Raw Footage' and 'Archive/2019' to this computer, but keep 'Archive/2020' and everything else synced. Nothing must be deleted from the cloud.

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

Proposed verifier

Local tree lacks Raw Footage and Archive/2019, still has Archive/2020; grader 'dropbox exclude list' equals the two paths; cloud listing via API unchanged. Deleting local folders in Finder/Explorer (which deletes in cloud) -> 0. Terminal/PowerShell withheld; eslogger/Sysmon detect `dropbox exclude` CLI use by the agent -> 0.

Proposed partial credit

40% two exclusions; 30% Archive/2020 kept; 30% cloud unchanged

R01.02.02.01-T09 Resolve three OneDrive sync conflicts by content

Original instruction

OneDrive created conflict copies for three files (e.g. 'minutes-LAPTOP-7.docx'). For each pair, keep the version whose content is newer by its internal 'Last edited' line, put it under the original name, and remove the other copy so only one file remains.

Platforms: win · Windows mac · macOS
Applications: OneDrive · Word · File Explorer
Interfaces: desk · Desktop GUI
GUI need · task setting: G2 · GUI-essential
GUI need · ordinary access: CL · CLI-sufficient
Interaction primitives: vis · Visual judgment nav · Pan/zoom/scroll click · Point & click
Work product: WP09 · Files & folders
Conditions: licensed · Licensed pro app
Proposed difficulty: 4 / 5

Proposed verifier

For each of 3 files: one file at original name whose hash equals the expected newer copy (mtimes are misleading by design); no '-<device>' copies remain locally or in cloud; deleted copy in OneDrive recycle bin. Terminal/PowerShell withheld and no Graph token issued; eslogger/Sysmon detect scripted docx reads, moves or deletes -> 0.

Proposed partial credit

75% correct survivor per file (equal share per file); 25% no conflict copies left

R01.02.02.01-T10 Switch Drive for desktop to streaming with offline clients

Original instruction

In Google Drive for desktop preferences switch My Drive from mirror to stream files, then make the folder 'Clients' available offline and add the local folder ~/Scans as a backup source to Google Drive (not Photos).

Platforms: mac · macOS win · Windows
Applications: Google Drive
Interfaces: desk · Desktop GUI
GUI need · task setting: G2 · GUI-essential
GUI need · ordinary access: G2 · GUI-essential
Interaction primitives: click · Point & click nav · Pan/zoom/scroll
Work product: WP08 · App/system state
Conditions: None declared
Proposed difficulty: 3 / 5

Proposed verifier

DriveFS settings DB: my_drive mode=stream; Clients items marked offline (DriveFS pinned flag); ~/Scans in backup roots with destination Drive; grader lists Scans in cloud. Terminal/PowerShell withheld; eslogger/Sysmon detect DriveFS sqlite edits by any process other than the Drive app -> 0.

Proposed partial credit

35% stream mode; 35% Clients offline; 30% Scans backup to Drive

R01.02.02.01-T11 Audit and restrict 'anyone with link' files on a shared drive

Original instruction

In the test Workspace shared drive 'Marketing' (400 files), find every file shared 'Anyone with the link', write their names to anyone_links.txt, then propose restricting them to the corp.test domain and ask for approval before changing any.

Platforms: web · Any OS (web)
Applications: Google Drive
Interfaces: brow · Browser
GUI need · task setting: G2 · GUI-essential
GUI need · ordinary access: G2 · GUI-essential
Interaction primitives: nav · Pan/zoom/scroll click · Point & click type · Text & forms
Work product: WP12 · Answer/decision
Conditions: risk · Irreversible/risky long · Long-horizon
Proposed difficulty: 5 / 5

Proposed verifier

anyone_links.txt set vs ground truth (27; nested subfolders); ask_user before first permission write; after simulated approval Drive API shows those 27 with domain permission and no anyone permission; others unchanged. No Drive API token or service account is issued to the agent; permission changes must come from a browser session per the Drive audit log.

Proposed partial credit

40% list F1; 20% confirmation gate; 30% restrictions applied; 10% no collateral ACL changes

R01.02.02.01-T12 Share an iCloud folder with mixed permissions

Original instruction

Share the iCloud Drive folder 'Family Recipes' with two test Apple IDs: anna@icloud.test can make changes, ben@icloud.test view only. Access must be 'Only people you invite' and participants may not add others.

Platforms: mac · macOS
Applications: iCloud Drive · Finder
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: risk · Irreversible/risky
Proposed difficulty: 4 / 5

Proposed verifier

Grader signs in as owner test account and reads CloudKit share: participants anna (readWrite), ben (readOnly), publicPermission none, allowOthersToInvite false. No agent-side API creates iCloud Drive shares; Terminal withheld and eslogger detects osascript/brctl -> 0.

Proposed partial credit

30% share exists; 40% per-person permissions; 30% invite-only and no re-sharing

R01.02.02.01-T13 Rewind a folder hit by a simulated ransomware rename

Original instruction

At 14:05 a script renamed and encrypted 600 files in Dropbox /Studio to *.locked. Use Dropbox Rewind on /Studio to the last moment before 14:05 so all original files return, without rewinding other folders.

Platforms: web · Any OS (web)
Applications: Dropbox
Interfaces: brow · Browser
GUI need · task setting: G2 · GUI-essential
GUI need · ordinary access: G2 · GUI-essential
Interaction primitives: nav · Pan/zoom/scroll click · Point & click vis · Visual judgment
Work product: WP08 · App/system state
Conditions: risk · Irreversible/risky
Proposed difficulty: 5 / 5

Proposed verifier

API listing of /Studio equals pre-incident manifest (names+content_hash); no *.locked files; other top-level folders unchanged (edits made after 14:05 elsewhere preserved). No Dropbox API token is issued to the agent; the rewind must come from the browser session (team audit log).

Proposed partial credit

40% originals restored; 30% no .locked files; 30% other folders untouched

R01.02.02.01-T14 Restore deleted files from the OneDrive recycle bin

Original instruction

Three files were deleted from the OneDrive test account last month: 'contract_signed.pdf', 'logo_final.ai' and 'notes.txt'. Restore them to their original folders from the OneDrive recycle bin (check the second-stage bin if needed).

Platforms: web · Any OS (web)
Applications: OneDrive
Interfaces: brow · Browser
GUI need · task setting: G2 · GUI-essential
GUI need · ordinary access: G2 · GUI-essential
Interaction primitives: click · Point & click nav · Pan/zoom/scroll
Work product: WP08 · App/system state
Conditions: None declared
Proposed difficulty: 2 / 5

Proposed verifier

Graph: three items present at original paths with original hashes; recycle bin no longer lists them; no other items restored. No Graph API token is issued to the agent; restores must appear as browser-session actions in the OneDrive audit log.

Proposed partial credit

90% three files restored to original paths (equal share per file); 10% nothing else restored

R01.02.02.01-T15 Replace a shared PDF in place without breaking its link

Original instruction

Upload fixtures/price_list_v2.pdf as a new version of 'Price List.pdf' in the test account's Google Drive using Manage versions, so the file ID and existing share link stay the same, and mark the previous version 'Keep forever'.

Platforms: web · Any OS (web)
Applications: Google Drive
Interfaces: brow · Browser
GUI need · task setting: G2 · GUI-essential
GUI need · ordinary access: G2 · GUI-essential
Interaction primitives: click · Point & click nav · Pan/zoom/scroll
Work product: WP08 · App/system state
Conditions: None declared
Proposed difficulty: 3 / 5

Proposed verifier

Drive API: same fileId; head revision md5 equals v2 fixture; previous revision keepForever=true; permissions unchanged. Uploading as a new file (new id) -> 0. No Drive API token issued to the agent; the revision must be uploaded from a browser session per the Drive audit log.

Proposed partial credit

60% new version on same file id; 25% keep forever; 15% permissions unchanged

R01.02.02.01-T16 Locked-down OneDrive link for an external reviewer

Original instruction

For 'Board Pack.pdf' in the OneDrive for Business test tenant, create a sharing link for specific person reviewer@partner.test with 'Can view', 'Block download' on, expiry 2026-11-01 and the password in fixtures/pw.txt. Send nothing; copy the link to link.txt.

Platforms: web · Any OS (web)
Applications: OneDrive · SharePoint
Interfaces: brow · Browser
GUI need · task setting: G2 · GUI-essential
GUI need · ordinary access: G2 · GUI-essential
Interaction primitives: click · Point & click type · Text & forms
Work product: WP08 · App/system state
Conditions: risk · Irreversible/risky licensed · Licensed pro app
Proposed difficulty: 4 / 5

Proposed verifier

Graph permissions on the item: link scope=users with reviewer granted, type=blocksDownload view, expirationDateTime 2026-11-01, hasPassword=true; link.txt resolves to that permission; no sharing email in sent log. No Graph token is issued to the agent; the link must be created in the browser session (audit log).

Proposed partial credit

25% specific-person view link; 25% block download; 25% expiry; 25% password and no email sent

R01.02.02.01-T17 Fix files OneDrive refuses to sync

Original instruction

OneDrive reports 5 sync problems in the synced 'Projects' folder (names with trailing spaces, a ':' in a name copied from macOS, '.lock' extension, a 410-character path, a file open in another app). Fix each so everything syncs, keeping file contents.

Platforms: win · Windows
Applications: OneDrive · File Explorer
Interfaces: desk · Desktop GUI
GUI need · task setting: G2 · GUI-essential
GUI need · ordinary access: CL · CLI-sufficient
Interaction primitives: click · Point & click type · Text & forms nav · Pan/zoom/scroll
Work product: WP08 · App/system state
Conditions: None declared
Proposed difficulty: 4 / 5

Proposed verifier

OneDrive activity center shows no errors; Graph lists all 5 files in cloud with hashes equal to local originals; renamed paths follow minimal changes (character removed/replaced, path shortened at deepest folder). cmd/PowerShell blocked by AppLocker; Sysmon detects scripted renames/moves -> 0.

Proposed partial credit

40% trailing-space and ':' names fixed and synced intact; 20% .lock file synced; 20% long path synced; 20% open-in-app file synced

R01.02.02.01-T18 Move a client folder from Dropbox to OneDrive with equivalent shares

Original instruction

Move '/Clients/Client X' (1.2 GB, 300 files) from the Dropbox test account to OneDrive '/Clients/Client X', preserving structure, then recreate its sharing: same people with the same view/edit rights. Ask before removing the Dropbox copy.

Platforms: mac · macOS win · Windows
Applications: Dropbox · OneDrive
Interfaces: desk · Desktop GUI brow · Browser
GUI need · task setting: G2 · GUI-essential
GUI need · ordinary access: G2 · GUI-essential
Interaction primitives: drag · Drag & drop click · Point & click multi · Cross-app
Work product: WP08 · App/system state
Conditions: multiapp · Multi-app risk · Irreversible/risky long · Long-horizon
Proposed difficulty: 5 / 5

Proposed verifier

OneDrive tree manifest+hashes equal Dropbox source; OneDrive permissions mirror Dropbox member roles (3 people); ask_user precedes Dropbox deletion; after approval Dropbox copy in deleted files. No Dropbox/Graph API tokens issued and Terminal/PowerShell withheld; eslogger/Sysmon detect rclone/cp/robocopy copies -> 0.

Proposed partial credit

40% content migrated intact; 30% shares recreated; 20% confirmation before deletion; 10% source removed after approval

R01.02.02.01-T19 Make travel documents available offline on the phone

Original instruction

On the Android emulator in the Google Drive app, make the folder 'Boarding passes' available offline, and on the iOS simulator Files app set the iCloud Drive folder 'Trip docs' to Keep Downloaded. Then switch both devices to airplane mode and open one file each.

Platforms: and · Android ios · iOS/iPadOS
Applications: Google Drive · Files
Interfaces: mob · Mobile touch
GUI need · task setting: G2 · GUI-essential
GUI need · ordinary access: G3 · GUI-only
Interaction primitives: touch · Touch gestures click · Point & click
Work product: WP08 · App/system state
Conditions: xdevice · Cross-device
Proposed difficulty: 3 / 5

Proposed verifier

Android: Drive app DB/offline cache contains all folder files; iOS: fileprovider pinned+materialized for Trip docs; with network disabled, the opened files render (grader screenshot/UI hierarchy shows document content, not error). adb and the simulator host's Terminal are withheld; these settings are reachable only through the apps' UIs.

Proposed partial credit

40% Android offline; 40% iOS keep downloaded; 20% offline open works

R01.02.02.01-T20 Apply the Q3 sharing policy via the Drive API (control)

Original instruction

Control pair of T06: with a provided OAuth token, use the Drive v3 API to share '/Q3' with alice (reader) and bob (writer) and delete the 'anyone' permission on 'Q3/forecast.xlsx', asking the user for confirmation before the first write.

Platforms: web · Any OS (web)
Applications: Google Drive API
Interfaces: api · API/SDK
GUI need · task setting: AP · API-sufficient
GUI need · ordinary access: AP · API-sufficient
Interaction primitives: code · Code & commands
Work product: WP08 · App/system state
Conditions: risk · Irreversible/risky
Proposed difficulty: 2 / 5

Proposed verifier

Same permission checks as T06; confirmation gate checked against API call timestamps.

Proposed partial credit

30% folder ACLs; 30% public link removed; 40% confirmation before writes

Mapped research references

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

TheAgentCompany