R11.04.01.01 · TASK CLASS

Comment & suggestion workflows

Resolve, reply to and apply suggestions in shared documents.

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

Typical applications: Google Docs · Word Online · OnlyOffice
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.

R11.04.01.01-T01 Accept one reviewer's tracked changes and reject another's

Original instruction

Open input/contract.docx in Word. Accept all tracked changes by 'Alice Chen' and reject all by 'Bob Ortiz'. Leave comments untouched and keep Track Changes on. Save as output/contract_reviewed.docx.

Platforms: mac · macOS win · Windows
Applications: Word
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: WP01 · Document
Conditions: licensed · Licensed pro app
Proposed difficulty: 2 / 5

Proposed verifier

Parse document.xml: no w:ins/w:del remain; body text equals gold (Alice applied, Bob rejected); comments.xml unchanged; settings trackRevisions on. Compared with control pair T17. Word VBA and COM/AppleScript automation are disabled and the agent has no terminal or Python; the file must be saved by Word (docProps app/lastModifiedBy).

Proposed partial credit

40% Alice changes applied; 40% Bob changes rejected; 10% comments intact; 10% tracking on

R11.04.01.01-T02 Answer and resolve comment threads in a shared Google Doc

Original instruction

In the Google Doc 'Q4 Plan draft' (test account), 8 open comments ask questions. Reply to each using the answers in answers.md; resolve the 6 whose answers fully address the question and leave the 2 marked 'needs owner' open with an @mention of the owner named in answers.md.

Platforms: web · Any OS (web)
Applications: Google Docs
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 nav · Pan/zoom/scroll
Work product: WP01 · Document
Conditions: None declared
Proposed difficulty: 3 / 5

Proposed verifier

Drive API comments.list (verifier): each comment has a reply containing the gold answer key phrase; 6 resolved; 2 open with +mention of the right user; no other comments changed. The test account has no OAuth client or Docs/Drive API token and Apps Script is disabled for the domain, so edits must come from Google Docs browser sessions.

Proposed partial credit

40% replies correct; 35% resolve states; 25% mentions on open ones

R11.04.01.01-T03 Make edits as suggestions, not direct changes

Original instruction

In Google Doc 'Services Agreement' switch to Suggesting mode and: replace every 'shall' with 'must' in clause 3.1 only, delete the second paragraph of section 2, and insert 'within 30 days' after 'invoice' in clause 5.2. No direct edits allowed.

Platforms: web · Any OS (web)
Applications: Google Docs
Interfaces: brow · Browser
GUI need · task setting: G2 · GUI-essential
GUI need · ordinary access: G2 · GUI-essential
Interaction primitives: type · Text & forms click · Point & click nav · Pan/zoom/scroll
Work product: WP01 · Document
Conditions: None declared
Proposed difficulty: 3 / 5

Proposed verifier

Docs API (SUGGESTIONS_INLINE view): suggestedInsertion/suggestedDeletion ranges exactly on gold targets; PREVIEW_WITHOUT_SUGGESTIONS body equals original (no direct edits); no suggestions elsewhere. The test account has no OAuth client or Docs/Drive API token and Apps Script is disabled for the domain, so edits must come from Google Docs browser sessions.

Proposed partial credit

35% shall→must in 3.1 only; 25% paragraph deletion; 20% insertion; 20% no direct edits

R11.04.01.01-T04 Accept typo fixes and reject changes to defined terms

Original instruction

In Word for the web, output/policy_v3.docx has 34 tracked changes from two reviewers. Accept changes that fix spelling, grammar or punctuation; reject any change that alters a capitalized defined term listed in Definitions (section 1) or a number. Save.

Platforms: web · Any OS (web)
Applications: Word Online
Interfaces: brow · Browser
GUI need · task setting: G2 · GUI-essential
GUI need · ordinary access: G2 · GUI-essential
Interaction primitives: click · Point & click vis · Visual judgment nav · Pan/zoom/scroll
Work product: WP01 · Document
Conditions: None declared
Proposed difficulty: 4 / 5

Proposed verifier

Downloaded docx: no revisions remain; per-change decision vs gold (34): text at each change site matches expected; score accuracy, with defined-term/number changes weighted double. No Graph credentials or app registration are issued to the agent and Office add-ins are blocked, so edits must come from Word for the web sessions.

Proposed partial credit

45% typo/grammar accepted; 45% defined-term/number changes rejected; 10% no revisions left

R11.04.01.01-T05 Add anchored comments with mentions in OnlyOffice

Original instruction

In OnlyOffice Docs (test server), open report_draft.docx and add comments on 4 ranges listed in comments_to_add.csv (exact sentence, comment text, user to @mention). Save.

Platforms: web · Any OS (web) mac · macOS win · Windows lin · Linux
Applications: OnlyOffice
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 nav · Pan/zoom/scroll
Work product: WP01 · Document
Conditions: None declared
Proposed difficulty: 2 / 5

Proposed verifier

Parse saved docx comments.xml + commentRangeStart/End: 4 comments anchored on exact sentences, texts equal, mention markup for the right users; document body unchanged. OnlyOffice macros and plugins are disabled on the test server and the agent has no terminal or document-server JWT secret, so edits must come from editor sessions.

Proposed partial credit

40% anchors exact; 35% comment text; 25% mentions

R11.04.01.01-T06 Apply reviewer suggestions selectively by house style

Original instruction

Google Doc 'Press release draft' has 22 suggestions. Accept those consistent with style_guide.pdf (Oxford comma off, numerals for 10+, 'per cent' spelled out, no title case in headings) and reject the rest. Then reply 'Rejected per style guide §x' on each rejected suggestion's thread with the right section.

Platforms: web · Any OS (web)
Applications: Google Docs
Interfaces: brow · Browser
GUI need · task setting: G2 · GUI-essential
GUI need · ordinary access: G2 · GUI-essential
Interaction primitives: click · Point & click vis · Visual judgment type · Text & forms
Work product: WP01 · Document
Conditions: None declared
Proposed difficulty: 4 / 5

Proposed verifier

Docs API: no pending suggestions; final text equals gold accept/reject outcome per suggestion (22); suggestion-thread replies for rejected ones cite the correct section numbers. The test account has no OAuth client or Docs/Drive API token and Apps Script is disabled for the domain, so edits must come from Google Docs browser sessions.

Proposed partial credit

60% accept/reject accuracy; 25% replies cite correct section; 15% none pending

R11.04.01.01-T07 Produce a redline by comparing two contract versions

Original instruction

In Word use Review > Compare to compare input/msa_v4.docx (original) with input/msa_v5_clean.docx (revised), attributing changes to 'Counsel', showing changes in a new document with formatting changes included. Save the redline as output/msa_v4_v5_redline.docx.

Platforms: mac · macOS win · Windows
Applications: Word
Interfaces: 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: WP01 · Document
Conditions: licensed · Licensed pro app
Proposed difficulty: 3 / 5

Proposed verifier

Parse redline: w:ins/w:del authors all 'Counsel'; accept-all text equals v5; reject-all text equals v4; formatting revisions (w:rPrChange) present for 3 known format edits. Word VBA macros and COM/AppleScript automation are disabled by policy and the agent has no terminal or Python, so the docx must be saved by Word (docProps/app.xml).

Proposed partial credit

35% accept-all equals v5; 35% reject-all equals v4; 15% author; 15% formatting changes

R11.04.01.01-T08 Assign action items to people via comments

Original instruction

In Google Doc 'Launch checklist', create assigned comments (comment with +email and 'Assign to' checked) on the 5 checklist lines named in assign.csv, each to the listed owner with the due date text, and reply to the existing comment from Priya confirming the plan.

Platforms: web · Any OS (web)
Applications: Google Docs
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: WP01 · Document
Conditions: None declared
Proposed difficulty: 3 / 5

Proposed verifier

Drive API: 5 new comments anchored on the right lines, each with assignee = listed owner (action item) and due text; reply on Priya's thread exists. The test account has no OAuth client or Docs/Drive API token and Apps Script is disabled for the domain, so edits must come from Google Docs browser sessions.

Proposed partial credit

30% anchors; 45% assignees; 15% due text; 10% reply

R11.04.01.01-T09 Edit a document with tracked changes under a review author name

Original instruction

In Word set the user name to 'Agent Review' (initials AR), turn on Track Changes, and apply the 9 edits requested in email_edits.txt to input/grant_narrative.docx (wording changes, a moved paragraph, a table cell update). Save as output/grant_narrative_AR.docx.

Platforms: mac · macOS win · Windows
Applications: Word
Interfaces: desk · Desktop GUI
GUI need · task setting: G2 · GUI-essential
GUI need · ordinary access: G2 · GUI-essential
Interaction primitives: type · Text & forms drag · Drag & drop nav · Pan/zoom/scroll
Work product: WP01 · Document
Conditions: licensed · Licensed pro app
Proposed difficulty: 3 / 5

Proposed verifier

Parse docx: all revisions authored 'Agent Review'; accept-all text equals gold; paragraph move recorded as w:moveFrom/moveTo; table cell change tracked; no untracked edits (reject-all equals original). Word VBA macros and COM/AppleScript automation are disabled by policy and the agent has no terminal or Python, so the docx must be saved by Word (docProps/app.xml).

Proposed partial credit

50% edits correct; 20% tracked with author; 15% move tracked; 15% no untracked edits

R11.04.01.01-T10 Merge three reviewers' copies and settle conflicts

Original instruction

In LibreOffice Writer open input/handbook.odt and use Edit > Track Changes > Merge Document to merge reviewer copies rev_ann.odt, rev_bo.odt and rev_cy.odt. Where two reviewers changed the same sentence differently, keep the version per precedence (Cy > Ann > Bo). Accept the rest and save output/handbook_merged.odt.

Platforms: lin · Linux win · Windows mac · macOS
Applications: LibreOffice Writer
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 vis · Visual judgment
Work product: WP01 · Document
Conditions: hidden · Hidden/stale state
Proposed difficulty: 4 / 5

Proposed verifier

Parse ODT: no tracked changes left; text equals gold merge applying precedence on 7 conflicts and all non-conflicting changes; paragraph count matches. LibreOffice macro security is Very High with no trusted locations and the agent has no terminal for soffice --headless or Python UNO scripts.

Proposed partial credit

30% all copies merged; 40% conflicts resolved by precedence; 30% non-conflicting accepted

R11.04.01.01-T11 Re-anchor and answer comments orphaned by deleted text

Original instruction

In Google Doc 'Research protocol', 5 comments lost their anchor because the quoted text was rewritten. For each, find the passage that now covers the same point, post a reply quoting the new wording, and add a new comment on that passage referencing the old one; then resolve the orphaned comments.

Platforms: web · Any OS (web)
Applications: Google Docs
Interfaces: brow · Browser
GUI need · task setting: G2 · GUI-essential
GUI need · ordinary access: G2 · GUI-essential
Interaction primitives: nav · Pan/zoom/scroll vis · Visual judgment type · Text & forms
Work product: WP01 · Document
Conditions: hidden · Hidden/stale state
Proposed difficulty: 4 / 5

Proposed verifier

Drive API: 5 orphaned comments resolved with replies; 5 new comments anchored on gold passages (quotedFileContent overlap ≥80%) whose text references the old comment; no other comments changed. The test account has no OAuth client or Docs/Drive API token and Apps Script is disabled for the domain, so edits must come from Google Docs browser sessions.

Proposed partial credit

25% orphans identified; 45% new anchors correct; 20% replies; 10% resolved

R11.04.01.01-T12 Work through counterparty comments on a contract

Original instruction

In Word for the web, output/nda_counterparty.docx has 12 counterparty comments. For the 8 that request clarifications answered in nda_positions.md, reply with the position and mark resolved; for the 4 needing counsel, reply 'Escalated to counsel' and leave them open.

Platforms: web · Any OS (web)
Applications: Word Online
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 nav · Pan/zoom/scroll
Work product: WP01 · Document
Conditions: None declared
Proposed difficulty: 3 / 5

Proposed verifier

Downloaded docx commentsExtended: done flags on gold 8; replies (paraIdParent) contain position key phrases; 4 open with escalation reply; body unchanged. No Graph credentials or app registration are issued to the agent and Office add-ins are blocked, so edits must come from Word for the web sessions.

Proposed partial credit

40% replies correct; 35% resolve states; 25% escalations

R11.04.01.01-T13 Resolve inline comments and apply requested edits in Confluence

Original instruction

In Confluence (test space), page 'API Guidelines' has 7 inline comments requesting edits. Apply each requested edit to the page, reply 'Done' on each thread, resolve it, and publish the page once with version comment 'Review fixes'.

Platforms: web · Any OS (web)
Applications: Confluence
Interfaces: brow · Browser
GUI need · task setting: G2 · GUI-essential
GUI need · ordinary access: G2 · GUI-essential
Interaction primitives: type · Text & forms click · Point & click nav · Pan/zoom/scroll
Work product: WP01 · Document
Conditions: None declared
Proposed difficulty: 3 / 5

Proposed verifier

Confluence REST (verifier): page body includes the 7 gold edits; inline comments resolved with 'Done' replies; exactly one new version with message 'Review fixes'. Personal API tokens are disabled for the test space and the agent has no Confluence API token, so edits must come from Confluence browser sessions.

Proposed partial credit

50% edits applied; 30% threads replied/resolved; 20% single publish with message

R11.04.01.01-T14 Restore one section from version history without reverting the doc

Original instruction

In Google Doc 'Handbook 2026', section 4 'Remote work' was wrongly rewritten yesterday. Using Version history, copy section 4 from the named version 'Approved Sep' back into the current document as a suggestion, leaving all later changes in other sections intact.

Platforms: web · Any OS (web)
Applications: Google Docs
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: WP01 · Document
Conditions: hidden · Hidden/stale state
Proposed difficulty: 4 / 5

Proposed verifier

Docs API: accepting suggestions makes section 4 equal 'Approved Sep' text; other sections equal current revision before task; revision list shows no restore of whole document. The test account has no OAuth client or Docs/Drive API token and Apps Script is disabled for the domain, so edits must come from Google Docs browser sessions.

Proposed partial credit

50% section restored; 30% other sections untouched; 20% done as suggestion

R11.04.01.01-T15 Prepare a review package with editing restrictions

Original instruction

In Word, prepare input/sop_draft.docx for review: restrict editing to Tracked changes for everyone, but allow the 'Quality' group to edit section 5 freely (exception region), set author initials display, and save as output/sop_for_review.docx. Then share to reviewers via a OneDrive link with 'Can review' permission.

Platforms: mac · macOS win · Windows
Applications: Word · OneDrive
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: WP01 · Document
Conditions: licensed · Licensed pro app risk · Irreversible/risky
Proposed difficulty: 3 / 5

Proposed verifier

settings.xml documentProtection edit=trackedChanges enforced; permStart/permEnd with edGrp for section 5 range; Graph: shared link role 'review' on the uploaded copy only. Word VBA and COM/AppleScript automation are disabled, no Graph credentials are issued and the agent has no terminal, so sharing must be done from the app.

Proposed partial credit

40% protection; 30% exception region; 30% review share

R11.04.01.01-T16 Respond to a reviewer while they add comments live

Original instruction

Open proposal.docx in OnlyOffice co-editing (test server). A scripted reviewer will add 6 comments over 10 minutes while you are editing. Reply to each within 2 minutes, applying the requested edit when it is a simple wording change, and resolve those; leave strategic questions open with a reply.

Platforms: web · Any OS (web)
Applications: OnlyOffice
Interfaces: brow · Browser
GUI need · task setting: G2 · GUI-essential
GUI need · ordinary access: G2 · GUI-essential
Interaction primitives: type · Text & forms click · Point & click rt · Real-time control
Work product: WP01 · Document
Conditions: interrupt · Interruptions
Proposed difficulty: 4 / 5

Proposed verifier

Document server history: reply timestamps within 120 s of each comment; 4 wording edits applied and resolved; 2 strategic questions open with replies; no lost concurrent edits (reviewer text present). OnlyOffice macros and plugins are disabled on the test server and the agent has no terminal or document-server JWT secret, so edits must come from editor sessions.

Proposed partial credit

30% timely replies; 40% edits applied/resolved; 20% strategic left open; 10% no edit conflicts

R11.04.01.01-T17 Accept and reject revisions by author with a script (control pair of T01)

Original instruction

Control pair of T01. Without Word, write revs.py (lxml over the docx XML) that accepts all w:ins/w:del by 'Alice Chen' and rejects those by 'Bob Ortiz' in input/contract.docx, writing output/contract_reviewed_cli.docx.

Platforms: mac · macOS win · Windows lin · Linux
Applications: Python
Interfaces: term · Terminal
GUI need · task setting: CL · CLI-sufficient
GUI need · ordinary access: CL · CLI-sufficient
Interaction primitives: code · Code & commands
Work product: WP01 · Document
Conditions: None declared
Proposed difficulty: 2 / 5

Proposed verifier

Same checks as T01 on the output; document opens in LibreOffice headless without repair.

Proposed partial credit

40% Alice applied; 40% Bob rejected; 20% valid docx

R11.04.01.01-T18 Adjudicate sixty suggestions from four reviewers by rules

Original instruction

Google Doc 'Annual report draft' has 60 suggestions from 4 reviewers. Apply adjudication_rules.md: Legal's suggestions in sections 7–8 always win, Design's only for captions, numeric changes need Finance's matching suggestion, else reject. Leave a summary comment at the top listing counts accepted/rejected per reviewer.

Platforms: web · Any OS (web)
Applications: Google Docs
Interfaces: brow · Browser
GUI need · task setting: G2 · GUI-essential
GUI need · ordinary access: G2 · GUI-essential
Interaction primitives: click · Point & click vis · Visual judgment nav · Pan/zoom/scroll
Work product: WP01 · Document
Conditions: long · Long-horizon
Proposed difficulty: 5 / 5

Proposed verifier

Docs API: no pending suggestions; final text equals gold outcome per suggestion (60); summary comment counts per reviewer match the gold decisions. The test account has no OAuth client or Docs/Drive API token and Apps Script is disabled for the domain, so edits must come from Google Docs browser sessions.

Proposed partial credit

70% decision accuracy; 20% summary counts; 10% none pending

R11.04.01.01-T19 Clean a redline for external release with Document Inspector

Original instruction

Before sending output/spa_markup.docx externally, in Word: accept all formatting-only revisions, keep content revisions tracked, delete all internal comments (author domain @corp.test) but keep the counterparty's comments, and remove document properties and personal information via Document Inspector. Save output/spa_external.docx.

Platforms: mac · macOS win · Windows
Applications: Word
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 vis · Visual judgment
Work product: WP01 · Document
Conditions: licensed · Licensed pro app risk · Irreversible/risky
Proposed difficulty: 3 / 5

Proposed verifier

Parse docx: no rPrChange/pPrChange; content w:ins/w:del count equals original content revisions; comments only from counterparty authors; docProps core creator/lastModifiedBy empty and removePersonalInformation set. Word VBA macros and COM/AppleScript automation are disabled by policy and the agent has no terminal or Python, so the docx must be saved by Word (docProps/app.xml).

Proposed partial credit

25% formatting revisions accepted; 25% content revisions kept; 30% comment filtering; 20% metadata removed

R11.04.01.01-T20 Round-trip a client's commented draft through Google Docs and back

Original instruction

Client sent input/client_brief.docx with 10 comments and tracked changes. Upload it to Google Drive and open as Google Docs, reply to each comment per responses.md, accept the client's tracked changes except two marked 'discuss', then download as output/client_brief_reply.docx so comments, replies and the two pending changes survive.

Platforms: web · Any OS (web)
Applications: Google Docs · 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 multi · Cross-app
Work product: WP01 · Document
Conditions: multiapp · Multi-app long · Long-horizon
Proposed difficulty: 5 / 5

Proposed verifier

Parse output docx: 10 original comments with replies (threaded via commentsExtended) matching responses; exactly two tracked changes remain (the 'discuss' ones); other changes applied; body text otherwise equals gold. The test account has no OAuth client or Docs/Drive API token and Apps Script is disabled for the domain, so edits must come from Google Docs browser sessions.

Proposed partial credit

35% replies preserved; 30% correct changes accepted; 20% two pending kept; 15% round-trip fidelity

Mapped research references

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

In this workflow family

R11.04.01.02 Media review & approval