Computer-Use Agent Taxonomy中文
R04.04.01.01 · TASK CLASS

Deck assembly from source content

Build a deck from a brief, document or data with the right structure and content.

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

Typical applications
PowerPoint · Keynote · Google Slides
Class GUI-need label
HY · Hybrid

Task proposals · 20

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

R04.04.01.01-T01 Turn an earnings memo into a six-slide deck on the corporate template

Original instruction

From input/earnings_memo.docx build output/q3_earnings.pptx on input/corp_template.pptx: title slide, 3 KPI slides (one per memo section, using the 'Title and Content' layout), 1 slide with a native column chart from the memo's revenue table, and a summary slide with the memo's 4 takeaways.

Platforms
mac · macOS win · Windows
Applications
PowerPoint · Word
Interfaces
desk · Desktop GUI
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
click · Point & click type · Text & forms multi · Cross-app
Work product
WP03 · Visual layout
Conditions
None declared
Proposed difficulty
2 / 5

Proposed verifier

python-pptx: 6 slides using template layouts (by layout name), titles equal memo headings, chart part is a native chart whose series values equal the memo table, takeaways text match (normalized); render via LibreOffice: no text box overflow (bbox within placeholder).

Proposed partial credit

20% slide count/layouts; 25% KPI slide text; 30% native chart values; 25% summary slide and no text overflow.

R04.04.01.01-T02 Build an agenda deck in Google Slides from a meeting doc

Original instruction

From Google Doc 'Offsite agenda' build Google Slides deck 'Offsite 2026' using the theme of 'Company template': a title slide and one slide per agenda item (5 items) with the item as title and owner, time slot and 3 bullet goals in the body, in agenda order.

Platforms
web · Any OS (web)
Applications
Google Slides · Google Docs
Interfaces
brow · Browser
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
click · Point & click type · Text & forms multi · Cross-app
Work product
WP03 · Visual layout
Conditions
None declared
Proposed difficulty
2 / 5

Proposed verifier

Slides API read: 6 slides, master/theme id equals template's, titles and body text equal agenda items (normalized), order correct; no empty placeholders.

Proposed partial credit

20% theme applied; 50% five item slides correct (pro-rata); 20% order; 10% no empty placeholders.

R04.04.01.01-T03 Assemble a Keynote launch deck from a brief and an image folder

Original instruction

In Keynote, build output/launch.key from input/launch_brief.md and the 9 images in input/product_shots/ using theme input/brand.kth: 8 slides as outlined in the brief, each product image placed in the master's media placeholder (not floating), headings in title placeholders, and presenter notes from the brief's 'Say' lines.

Platforms
mac · macOS
Applications
Keynote
Interfaces
desk · Desktop GUI
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
click · Point & click drag · Drag & drop type · Text & forms
Work product
WP03 · Visual layout
Conditions
None declared
Proposed difficulty
3 / 5

Proposed verifier

Grader AppleScript export to .pptx + slide inspection: 8 slides on the brand theme masters, images in media placeholders (matched by perceptual hash), titles and notes equal brief text; exported PDF render: no overflowing text.

Proposed partial credit

20% theme/masters; 30% images in placeholders; 25% titles/body text; 25% presenter notes and no overflow.

R04.04.01.01-T04 Import a Word outline into PowerPoint and map it to template layouts

Original instruction

Create output/policy_training.pptx from input/policy_outline.docx: each Heading 1 becomes a slide title, Heading 2s become first-level bullets and Heading 3s second-level. Apply input/hr_template.potx so slides use its 'Title and Content' layout, and make the first slide 'Title Slide'. Remove any blank slides the import leaves.

Platforms
mac · macOS win · Windows
Applications
PowerPoint · Word
Interfaces
desk · Desktop GUI
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
click · Point & click multi · Cross-app
Work product
WP03 · Visual layout
Conditions
None declared
Proposed difficulty
3 / 5

Proposed verifier

Slide count = number of Heading 1s; titles and bullet text/levels equal the outline; layouts from hr_template by name; first slide 'Title Slide'; no empty slides; template theme fonts present.

Proposed partial credit

25% slide count/titles; 35% bullet levels; 25% layouts/template; 15% no blanks.

R04.04.01.01-T05 Add a native chart and table from a sales workbook to a deck

Original instruction

Extend input/regional_review.pptx with two slides after slide 4: a clustered column chart of Q3 revenue by region from input/q3_sales.xlsx (native PowerPoint chart, editable data) and a 5×4 table with region, revenue, growth %, margin % formatted as in the workbook. Use the deck's 'Title Only' layout.

Platforms
mac · macOS win · Windows
Applications
PowerPoint · Excel
Interfaces
desk · Desktop GUI
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
click · Point & click multi · Cross-app type · Text & forms
Work product
WP03 · Visual layout
Conditions
multiapp · Multi-app
Proposed difficulty
3 / 5

Proposed verifier

New slides at positions 5 and 6 with layout 'Title Only'; chart is a graphicFrame chart (not picture) with embedded workbook whose values equal q3_sales; table cell text equals formatted workbook values; other slides unchanged (hash).

Proposed partial credit

15% placement/layout; 40% native chart values; 35% table values and formats; 10% other slides unchanged.

R04.04.01.01-T06 Turn a research paper into a 12-slide conference talk

Original instruction

From input/paper.pdf build output/talk.pptx: 12 slides (title, motivation, method ×3, results ×4, ablation, limitations, conclusion). Each results slide shows one of the paper's figures 2–5 cropped from the PDF at native resolution with its figure number in the caption. Put a 2–4 sentence script in each slide's notes.

Platforms
mac · macOS win · Windows
Applications
PowerPoint · Acrobat
Interfaces
desk · Desktop GUI
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
click · Point & click multi · Cross-app vis · Visual judgment
Work product
WP03 · Visual layout
Conditions
multiapp · Multi-app
Proposed difficulty
3 / 5

Proposed verifier

12 slides; figures 2–5 matched to PDF crops by perceptual hash (≥0.9) and ≥150 dpi effective; captions contain 'Figure N'; notes 2–4 sentences each; section titles mapped to paper sections. Narrative quality scored by an LLM rubric (20%) since no deterministic check exists.

Proposed partial credit

20% structure/count; 35% correct figures and captions; 25% notes; 20% LLM-rubric faithfulness.

R04.04.01.01-T07 Fill one QBR slide per account from a CRM export in Google Slides

Original instruction

Deck 'QBR template' has a 2-slide pattern (account summary, pipeline). Using input/crm_export.csv, create a deck 'QBR Oct' with that pattern duplicated for each of 8 accounts, filling ARR, renewal date, health and top 3 open opportunities, in descending ARR order. Delete the template pattern slides.

Platforms
web · Any OS (web)
Applications
Google Slides · Google Sheets
Interfaces
brow · Browser
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
click · Point & click type · Text & forms multi · Cross-app
Work product
WP03 · Visual layout
Conditions
None declared
Proposed difficulty
3 / 5

Proposed verifier

Slides API: 16 slides, pairs ordered by ARR; text fields equal CSV values (formatted per template); no placeholder tokens like '{{ARR}}' remain; template slides absent.

Proposed partial credit

20% slide count and order; 60% account fields correct (pro-rata); 20% no leftover tokens/template slides.

R04.04.01.01-T08 Build a chronological photo deck with captions from a CSV

Original instruction

In Keynote, create output/field_trip.key with the 24 photos in input/trip_photos/: two photos per slide in capture-time order (EXIF DateTimeOriginal), each with a caption from input/captions.csv below it, photo frames equal size and top-aligned, and a title slide 'Field Trip 2026'.

Platforms
mac · macOS
Applications
Keynote · Photos
Interfaces
desk · Desktop GUI
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
drag · Drag & drop geom · Precise geometry type · Text & forms
Work product
WP03 · Visual layout
Conditions
None declared
Proposed difficulty
3 / 5

Proposed verifier

Grader exports to .pptx: 13 slides; photo order equals EXIF order (perceptual hash match); captions matched to photos; per-slide image boxes equal size (±2 px) and top edges aligned (±2 px).

Proposed partial credit

15% title slide; 35% photo order; 30% captions; 20% sizing/alignment.

R04.04.01.01-T09 Structure a training deck into sections with a summary zoom

Original instruction

From input/sop_manual.docx (6 chapters) build output/sop_training.pptx: one PowerPoint section per chapter named after it with 3–5 slides each, and a Summary Zoom slide after the title that links to all 6 sections. Use input/ops_template.potx.

Platforms
win · Windows mac · macOS
Applications
PowerPoint · Word
Interfaces
desk · Desktop GUI
GUI need · task setting
G2 · GUI-essential
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
click · Point & click drag · Drag & drop multi · Cross-app
Work product
WP03 · Visual layout
Conditions
None declared
Proposed difficulty
3 / 5

Proposed verifier

presentation.xml sectionLst: 6 named sections in chapter order with 3–5 slides each; summary-zoom slide (p15/psez zoom objects) linking to all 6 section ids; template layouts used. python-pptx cannot author zoom objects. VBA, COM and AppleScript automation disabled, no Python in the VM; file must be saved by PowerPoint (app.xml).

Proposed partial credit

35% sections named and ordered; 20% slide counts per section; 35% summary zoom links; 10% template.

R04.04.01.01-T10 Assemble a board deck from a workbook, a memo and a dashboard image

Original instruction

Build output/board_nov.pptx on input/board_template.pptx from three sources: input/financials.xlsx (P&L table and a native line chart of monthly revenue), input/board_memo.docx (strategy bullets), and input/kpi_dashboard.png (read the 5 KPI values into a KPI tile slide as text). Cite the source on each slide footer.

Platforms
mac · macOS win · Windows
Applications
PowerPoint · Excel · Word
Interfaces
desk · Desktop GUI
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
multi · Cross-app vis · Visual judgment type · Text & forms
Work product
WP03 · Visual layout
Conditions
multiapp · Multi-app
Proposed difficulty
4 / 5

Proposed verifier

Slide plan per template; P&L table values equal workbook; native chart series equal monthly revenue; KPI tile text equals dashboard values (exact); memo bullets match; each slide footer names its source file.

Proposed partial credit

25% P&L table; 25% native chart; 25% KPI values from image; 25% memo bullets and source footers.

R04.04.01.01-T11 Build a case-study deck with region screenshots from a web page

Original instruction

From the sandbox page http://casestudies.local/acme-logistics, build output/acme_case.pptx with 10 slides following input/storyboard.md; three slides need screenshots of specific page regions (the metrics band, the architecture diagram, the quote card) captured at 2× scale and cropped to just that element.

Platforms
mac · macOS win · Windows
Applications
PowerPoint · Chrome
Interfaces
desk · Desktop GUI brow · Browser
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
HY · Hybrid
Interaction primitives
multi · Cross-app vis · Visual judgment geom · Precise geometry
Work product
WP03 · Visual layout
Conditions
multiapp · Multi-app
Proposed difficulty
4 / 5

Proposed verifier

10 slides per storyboard; slide text matches page copy for quoted sections; each screenshot image compared to the grader's element crops (SSIM ≥0.9, pixel dims ≥2× CSS size, crop IoU ≥0.9).

Proposed partial credit

30% storyboard structure and text; 45% three screenshots correct (equal thirds); 25% sizing/placement.

R04.04.01.01-T12 Build a Keynote pitch deck with a native chart and export to PPTX

Original instruction

Create output/pitch.key in Keynote from input/investor_brief.md: 10 slides on theme input/startup.kth masters, a native Keynote line chart of MRR from input/mrr.csv (24 months), presenter notes from the brief, then export output/pitch.pptx that keeps the chart editable.

Platforms
mac · macOS
Applications
Keynote
Interfaces
desk · Desktop GUI
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
click · Point & click type · Text & forms multi · Cross-app
Work product
WP03 · Visual layout
Conditions
None declared
Proposed difficulty
4 / 5

Proposed verifier

pitch.pptx: 10 slides; chart graphicFrame with 24 points equal mrr.csv (editable chart, not picture); titles/notes equal brief; Keynote file opens (grader AppleScript) with same slide count and theme name.

Proposed partial credit

25% slide plan; 35% chart values and editability; 20% notes; 20% both files present and consistent.

R04.04.01.01-T13 Generate one chart slide per survey question with the right chart type

Original instruction

From input/survey_results.xlsx (15 questions: 9 single-choice, 6 Likert), build output/survey_readout.pptx with one slide per question in questionnaire order: single-choice as sorted horizontal bars of %, Likert as 100% stacked bars with the 5 levels in scale order and fixed colors from input/palette.txt.

Platforms
mac · macOS win · Windows
Applications
PowerPoint · Excel
Interfaces
desk · Desktop GUI
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
click · Point & click multi · Cross-app type · Text & forms
Work product
WP03 · Visual layout
Conditions
None declared
Proposed difficulty
3 / 5

Proposed verifier

15 chart slides in order; chart type per question type; series values equal workbook %; Likert series order and colors exact; titles equal question text.

Proposed partial credit

15% order/titles; 35% chart types; 35% values; 15% Likert order and colors.

R04.04.01.01-T14 Condense an annual report into a 20-slide investor deck by storyboard

Original instruction

From input/annual_report_2025.pdf build output/investor_day.pptx on input/ir_template.pptx following input/storyboard.xlsx (20 rows: slide title, layout, required numbers, chart spec). Every number must trace to the report; charts must be native; put the report page reference in each slide's notes.

Platforms
mac · macOS win · Windows
Applications
PowerPoint · Acrobat · Excel
Interfaces
desk · Desktop GUI
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
multi · Cross-app type · Text & forms vis · Visual judgment
Work product
WP03 · Visual layout
Conditions
long · Long-horizon multiapp · Multi-app
Proposed difficulty
5 / 5

Proposed verifier

Per storyboard row: layout name, title text, required numbers present and equal report values, native charts with correct series; notes contain correct page refs; LibreOffice render: no overflow. Fully deterministic per-row checklist.

Proposed partial credit

20% layouts/titles; 40% numbers correct (pro-rata); 25% native charts; 15% page references.

R04.04.01.01-T15 Merge three team decks into one themed, de-duplicated deck

Original instruction

Combine input/eng.pptx, input/sales.pptx and input/ops.pptx into Google Slides deck 'All-hands Q4' using theme 'Company 2026': follow input/agenda.md order, drop 5 duplicate slides (same content in two decks, keep the newer), keep all speaker notes, and renumber slide footers.

Platforms
web · Any OS (web)
Applications
Google Slides · PowerPoint
Interfaces
brow · Browser
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
drag · Drag & drop click · Point & click multi · Cross-app
Work product
WP03 · Visual layout
Conditions
long · Long-horizon multiapp · Multi-app
Proposed difficulty
5 / 5

Proposed verifier

Slides API: slide order matches agenda mapping; 5 seeded duplicates removed with newer versions kept; all slides on the target theme masters; speaker notes text preserved per slide; footer numbers sequential.

Proposed partial credit

30% order; 25% duplicates handled; 20% theme applied; 25% notes kept and numbering.

R04.04.01.01-T16 Build a decisions deck from a meeting transcript and whiteboard photo

Original instruction

From input/meeting_transcript.vtt and input/whiteboard.jpg build output/decisions.pptx: one slide of decisions, one of action items as a table (owner, task, due date) and one slide with the whiteboard photo straightened and cropped to the board. Only include items explicitly agreed in the transcript.

Platforms
mac · macOS win · Windows
Applications
PowerPoint
Interfaces
desk · Desktop GUI
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
vis · Visual judgment type · Text & forms geom · Precise geometry
Work product
WP03 · Visual layout
Conditions
None declared
Proposed difficulty
4 / 5

Proposed verifier

Decisions and action items matched to key (owner/date exact, task fuzzy ≥0.85); no unagreed items (precision); whiteboard image: perspective-corrected crop compared to reference (SSIM ≥0.85) via PowerPoint crop/picture edits.

Proposed partial credit

35% action items table; 25% decisions; 15% no extra items; 25% whiteboard crop.

R04.04.01.01-T17 Control pair of T01: generate the earnings deck with python-pptx

Original instruction

Control pair of T01. With python-pptx from the terminal, generate output/q3_earnings_cli.pptx from input/earnings_memo.docx and input/corp_template.pptx with the same 6-slide structure, native chart and summary.

Platforms
mac · macOS win · Windows lin · Linux
Applications
python-pptx · python-docx
Interfaces
term · Terminal
GUI need · task setting
CL · CLI-sufficient
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
code · Code & commands
Work product
WP03 · Visual layout
Conditions
None declared
Proposed difficulty
2 / 5

Proposed verifier

Same checker as T01, compared with the GUI attempt; overflow checked in LibreOffice render.

Proposed partial credit

20% slides/layouts; 25% KPI text; 30% chart values; 25% summary and no overflow.

R04.04.01.01-T18 Build a lecture deck with native editable equations from a chapter

Original instruction

From input/ch4_thermo.epub build output/lecture4.pptx on input/course_template.pptx: 14 slides following the chapter's subsections, with the 6 numbered equations re-typed as native PowerPoint equations (not images), each next to its explanation, and one worked example slide.

Platforms
mac · macOS win · Windows
Applications
PowerPoint
Interfaces
desk · Desktop GUI
GUI need · task setting
G2 · GUI-essential
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
type · Text & forms click · Point & click key · Shortcuts
Work product
WP03 · Visual layout
Conditions
None declared
Proposed difficulty
4 / 5

Proposed verifier

Slide count and titles follow subsections; 6 a14:m OMML equation objects whose math trees equal reference (normalized); no picture of an equation; worked example values equal chapter. python-pptx cannot author OMML math. VBA, COM and AppleScript automation disabled, no Python in the VM; file must be saved by PowerPoint (app.xml).

Proposed partial credit

20% structure; 50% six equations correct (pro-rata); 15% no equation images; 15% worked example.

R04.04.01.01-T19 Refresh an 18-slide weekly KPI deck from new data

Original instruction

Update input/kpi_week41.pptx to week 42 using input/kpis_week42.xlsx: change all 40 numbers and 6 native charts, flip commentary bullets whose thresholds (input/rules.md) are now crossed, update dates in titles and footers, and keep layout and formatting unchanged. Save output/kpi_week42.pptx.

Platforms
mac · macOS win · Windows
Applications
PowerPoint · Excel
Interfaces
desk · Desktop GUI
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
CL · CLI-sufficient
Interaction primitives
click · Point & click type · Text & forms multi · Cross-app
Work product
WP03 · Visual layout
Conditions
long · Long-horizon hidden · Hidden/stale state
Proposed difficulty
5 / 5

Proposed verifier

All 40 seeded number locations equal week-42 values; chart caches and embedded workbooks equal new data; commentary bullets match rule outcomes; no 'Week 41' strings remain; shape geometry and fonts unchanged vs input.

Proposed partial credit

40% numbers (pro-rata); 25% charts; 20% rule-driven bullets; 15% dates and unchanged formatting.

R04.04.01.01-T20 Create a speaker-profile deck from form responses and Drive photos

Original instruction

Sheet 'Speaker submissions' (14 rows) holds name, talk title, bio and a Drive photo link. Build Google Slides deck 'Summit speakers' with one slide per speaker using layout 'Speaker' of 'Summit template': photo in the image placeholder cropped to circle, name, title, bio ≤60 words, alphabetical by surname.

Platforms
web · Any OS (web)
Applications
Google Slides · Google Sheets · Google Drive
Interfaces
brow · Browser
GUI need · task setting
HY · Hybrid
GUI need · ordinary access
G2 · GUI-essential
Interaction primitives
click · Point & click multi · Cross-app type · Text & forms
Work product
WP03 · Visual layout
Conditions
None declared
Proposed difficulty
3 / 5

Proposed verifier

Slides API: 14 slides on layout 'Speaker', surname order; text equals sheet values (bio truncated to ≤60 words at sentence end); image source matches each Drive file id; image shape has ellipse crop.

Proposed partial credit

20% count/order; 40% text fields; 25% correct photos; 15% circle crop.

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