计算机使用智能体分类English
R01.10.01.04 · 任务小类

复杂网页控件

Complex web widgets

Operate date pickers, range sliders, autocomplete, hover menus, drag-and-drop lists and rich selects to exact values.

20 条本项目编写的提案 · 分类 v1.1 · 新增基准运行 0 次。下方验证方式与部分得分规则是设计,不是执行结果。

典型应用
Chrome · Safari · Edge
小类 GUI 需求标签
G3 · 仅 GUI

任务提案 · 20

展开查看完整原文与建议检查方式;两种界面均保留提案英文原文。

R01.10.01.04-T01 Set a dual-thumb price slider to exactly $35-$80

原始指令

Chrome is open at the replica store http://shop.local/c/backpacks with the price filter at $0-$250 (step $1). Set the dual-thumb slider to exactly $35 (min) and $80 (max) and apply the filter. The min/max text boxes are disabled; use the slider handles with mouse or arrow keys.

平台
mac · macOS win · Windows web · 任意系统(网页)
应用
Chrome · shop.local replica
交互表面
brow · 浏览器
GUI 需求 · 任务设定
G3 · 仅 GUI
GUI 需求 · 普通权限
G2 · GUI 必要
交互原语
drag · 拖放 geom · 精确几何
工作产物
WP08 · 应用 / 系统状态
条件
未声明
提案难度
2 / 5

建议验证方式

Request log: last GET /api/products has price_min=35&price_max=80 and URL state matches. Widget records only isTrusted pointer/key events; a programmatic value change scores 0. DevTools and javascript: URLs disabled by Chrome policy.

建议部分得分

40% min thumb = 35; 40% max thumb = 80; 20% filter applied (grid refreshed to the expected 14 items)

R01.10.01.04-T02 Pick a December-January stay in a two-month range picker

原始指令

In Edge, the replica hotel site stay.local opens a two-month range date picker on the current month (October 2026). Select check-in 2026-12-18 and check-out 2027-01-03 and press Search. The date inputs are read-only, so the picker is the only way to set them.

平台
win · Windows mac · macOS web · 任意系统(网页)
应用
Edge · stay.local replica
交互表面
brow · 浏览器
GUI 需求 · 任务设定
G3 · 仅 GUI
GUI 需求 · 普通权限
G2 · GUI 必要
交互原语
click · 指向与点击 nav · 平移 / 缩放 / 滚动
工作产物
WP08 · 应用 / 系统状态
条件
未声明
提案难度
2 / 5

建议验证方式

Search request has checkin=2026-12-18&checkout=2027-01-03 and the picker's internal range matches. Event log must show trusted clicks on month arrows and day cells; DevTools disabled by Edge policy.

建议部分得分

40% check-in correct; 40% check-out correct across the year boundary; 20% search submitted

R01.10.01.04-T03 Choose look-alike airports from an async autocomplete

原始指令

In Safari on the replica flight site fly.local, From and To are async autocompletes returning up to 8 suggestions per keystroke. Set From to San Jose, Costa Rica (SJO), not San Jose, California (SJC), and To to Portland, Maine (PWM), not Portland, Oregon (PDX), then press Search. Free text not picked from the list is rejected.

平台
mac · macOS web · 任意系统(网页)
应用
Safari · fly.local replica
交互表面
brow · 浏览器
GUI 需求 · 任务设定
G3 · 仅 GUI
GUI 需求 · 普通权限
G2 · GUI 必要
交互原语
type · 文字与表单 click · 指向与点击
工作产物
WP08 · 应用 / 系统状态
条件
未声明
提案难度
2 / 5

建议验证方式

Search request has origin=SJO&dest=PWM; hidden IATA fields are set only by a trusted suggestion click or Enter. Web Inspector disabled by MDM profile; URL-bar navigation to a results URL is logged and scores 0.

建议部分得分

45% origin SJO; 45% destination PWM; 10% search submitted

R01.10.01.04-T04 Navigate a hover-only mega-menu to a third-level filter

原始指令

On outdoor.local in Chrome, the top navigation uses hover-only flyouts with a 300 ms close delay and no click-to-open. Open Camping > Tents > 4-Person Tents and, from the same flyout, click the 'Freestanding' filter link in its fourth column. Do not use site search or type URLs.

平台
mac · macOS win · Windows web · 任意系统(网页)
应用
Chrome · outdoor.local replica
交互表面
brow · 浏览器
GUI 需求 · 任务设定
G3 · 仅 GUI
GUI 需求 · 普通权限
G2 · GUI 必要
交互原语
hover · 悬停与上下文 click · 指向与点击
工作产物
WP08 · 应用 / 系统状态
条件
未声明
提案难度
3 / 5

建议验证方式

Final URL /c/camping/tents/4-person?f=freestanding reached by a trusted click on that menu anchor (click log + referer). Search requests or typed URLs score 0; DevTools disabled by policy.

建议部分得分

30% Camping flyout opened; 30% Tents submenu reached; 40% correct final link clicked

R01.10.01.04-T05 Reorder a 12-item sortable backlog by drag-and-drop

原始指令

plan.local in Chrome shows a SortableJS backlog of 12 items in random order. Reorder it to match fixtures/priorities.txt (one title per line, top = highest) and click Save Order. Items have no keyboard reordering and no editable rank field.

平台
mac · macOS win · Windows web · 任意系统(网页)
应用
Chrome · plan.local replica
交互表面
brow · 浏览器
GUI 需求 · 任务设定
G3 · 仅 GUI
GUI 需求 · 普通权限
G2 · GUI 必要
交互原语
drag · 拖放
工作产物
WP08 · 应用 / 系统状态
条件
未声明
提案难度
3 / 5

建议验证方式

Backend rank list must equal the fixture order. Save must be the page's own POST preceded by trusted pointerdown/move/up telemetry for each moved item; DevTools disabled. Control pair: T19.

建议部分得分

80% x (1 - normalized Kendall tau distance to target); 20% saved through the page; full credit only for exact order

R01.10.01.04-T06 Move Kanban cards across columns without breaking WIP limits

原始指令

A Wekan board in Firefox has columns Backlog, Ready (WIP 4), Doing (WIP 3), Review, Done. Apply fixtures/standup_notes.md to swimlane 'Team B': move the 6 named cards to their new columns and positions. Sequence the moves so no column ever exceeds its WIP limit.

平台
mac · macOS win · Windows web · 任意系统(网页)
应用
Firefox · Wekan
交互表面
brow · 浏览器
GUI 需求 · 任务设定
G2 · GUI 必要
GUI 需求 · 普通权限
G2 · GUI 必要
交互原语
drag · 拖放
工作产物
WP08 · 应用 / 系统状态
条件
未声明
提案难度
3 / 5

建议验证方式

Wekan MongoDB: each card's listId, swimlaneId and sort match target. Activity log replay checks WIP limits at every step. Moves must be client drag method calls; REST API token withheld, DevTools disabled.

建议部分得分

60% cards in correct columns, equal shares over the 6; 25% order within columns; 15% no WIP violation in history

R01.10.01.04-T07 Select an exact country set in a grouped multi-select with chips

原始指令

The shipping-zones form at admin.local (Edge) uses a React-Select grouped, virtualized multi-select (240 options by continent). Make zone 'EU-Plus' contain exactly the 9 countries in fixtures/eu_plus.txt, including Aland Islands and Curacao (listed with diacritics), remove the 3 preselected chips not on the list, and save.

平台
win · Windows mac · macOS web · 任意系统(网页)
应用
Edge · admin.local replica
交互表面
brow · 浏览器
GUI 需求 · 任务设定
G3 · 仅 GUI
GUI 需求 · 普通权限
G2 · GUI 必要
交互原语
type · 文字与表单 click · 指向与点击 nav · 平移 / 缩放 / 滚动
工作产物
WP08 · 应用 / 系统状态
条件
未声明
提案难度
3 / 5

建议验证方式

DB zone_countries set equals fixture ISO codes. Option picks and chip removals must be trusted clicks/keys. Admin REST endpoint needs a CSRF token rendered in the form; DevTools disabled.

建议部分得分

80% x Jaccard(selected, target); 20% saved with no extra countries

R01.10.01.04-T08 Schedule a webinar with a spinner time picker and time-zone select

原始指令

In Chrome, the form at events.local has a calendar date picker, a 15-minute spinner time picker (hour/minute/AM-PM wheels), a searchable time-zone dropdown and a duration slider. Schedule 'Q4 Partner Briefing' for 2026-11-03 at 2:45 PM America/Sao_Paulo, 75 minutes long, and save it as a draft.

平台
mac · macOS win · Windows web · 任意系统(网页)
应用
Chrome · events.local replica
交互表面
brow · 浏览器
GUI 需求 · 任务设定
G3 · 仅 GUI
GUI 需求 · 普通权限
G2 · GUI 必要
交互原语
drag · 拖放 click · 指向与点击 type · 文字与表单
工作产物
WP08 · 应用 / 系统状态
条件
未声明
提案难度
3 / 5

建议验证方式

events row: start_utc=2026-11-03T17:45Z, tz=America/Sao_Paulo, duration=75, status=draft. Wheels and slider must change via trusted wheel/drag/key events; DevTools disabled.

建议部分得分

25% date; 25% time; 25% time zone; 25% duration + draft saved

R01.10.01.04-T09 Match two brand colors with an HSV picker without hex entry

原始指令

The theme editor at brand.local (Chrome) has an HSV saturation/value square and a hue slider; the hex field is read-only. Set Primary to match the swatch fixtures/brand_primary.png and Accent to fixtures/brand_accent.png, then Publish.

平台
mac · macOS win · Windows web · 任意系统(网页)
应用
Chrome · brand.local replica
交互表面
brow · 浏览器
GUI 需求 · 任务设定
G3 · 仅 GUI
GUI 需求 · 普通权限
G2 · GUI 必要
交互原语
drag · 拖放 geom · 精确几何 vis · 视觉判断
工作产物
WP08 · 应用 / 系统状态
条件
未声明
提案难度
4 / 5

建议验证方式

Published theme colors vs swatch sRGB: full credit at dE2000 <= 1.0, zero at >= 5.0, linear between. Trusted drags on square and hue slider required; paste into the picker ignored; DevTools disabled.

建议部分得分

45% Primary by dE2000; 45% Accent by dE2000; 10% published

R01.10.01.04-T10 Enter three addresses through lazy cascading country-region-city selects

原始指令

The address book at ship.local (Safari) uses cascading selects Country > Region > City; each list loads after its parent changes and a parent change clears its children. Add the 3 addresses in fixtures/addresses.csv (Japan, Brazil, Canada) with street lines and postal codes, saving each one.

平台
mac · macOS web · 任意系统(网页)
应用
Safari · ship.local replica
交互表面
brow · 浏览器
GUI 需求 · 任务设定
G3 · 仅 GUI
GUI 需求 · 普通权限
G2 · GUI 必要
交互原语
click · 指向与点击 type · 文字与表单
工作产物
WP08 · 应用 / 系统状态
条件
未声明
提案难度
3 / 5

建议验证方式

addresses rows equal fixture fields with region/city IDs from the lists, not free text. Child-list XHRs must follow trusted parent selections; direct API posts flagged. Web Inspector disabled via MDM.

建议部分得分

90% three addresses correct, equal shares (two-thirds of each for country/region/city, one-third for other fields); 10% no stray partial rows

R01.10.01.04-T11 Grant a role exact permissions in a lazy tri-state checkbox tree

原始指令

In the admin console iam.local (Edge), role 'Auditor' uses a checkbox tree of about 180 permissions; branches load on expand and parent boxes are tri-state. Grant exactly the permissions in fixtures/auditor_perms.txt (read scopes plus two export actions) and nothing else, then save. Do not use 'Select all'.

平台
win · Windows mac · macOS web · 任意系统(网页)
应用
Edge · iam.local replica
交互表面
brow · 浏览器
GUI 需求 · 任务设定
G3 · 仅 GUI
GUI 需求 · 普通权限
G2 · GUI 必要
交互原语
click · 指向与点击 nav · 平移 / 缩放 / 滚动
工作产物
WP08 · 应用 / 系统状态
条件
risk · 高风险 / 不可逆
提案难度
4 / 5

建议验证方式

role_permissions set vs fixture (F1). Any write/delete scope granted caps score at 30%. Expansion and checks must be trusted clicks; role-import endpoint disabled; DevTools blocked by policy.

建议部分得分

70% x F1 of granted vs target; 20% no write/delete scope; 10% saved

R01.10.01.04-T12 Build a saved view in a virtualized data grid by dragging columns

原始指令

An AG Grid table at ops.local (Chrome) lists 5,000 orders. Create saved view 'Late EU': drag Region into the row-group panel, reorder columns per fixtures/view_spec.json, resize Customer to 220 px, pin Order ID left, multi-sort Ship Date desc then Total desc, filter Status = Late, and save the view.

平台
mac · macOS win · Windows web · 任意系统(网页)
应用
Chrome · AG Grid app (ops.local)
交互表面
brow · 浏览器
GUI 需求 · 任务设定
G3 · 仅 GUI
GUI 需求 · 普通权限
G2 · GUI 必要
交互原语
drag · 拖放 geom · 精确几何 click · 指向与点击
工作产物
WP08 · 应用 / 系统状态
条件
未声明
提案难度
5 / 5

建议验证方式

Saved view JSON (columnState, rowGroup, sortModel, filterModel) diffed against spec; width 220 +/- 5 px. Column state must change via trusted header drags; grid API object removed from window, DevTools disabled.

建议部分得分

20% row group; 40% column order, width and pin; 20% sort model; 20% filter and view saved

R01.10.01.04-T13 Move 18 of 60 names across a dual listbox with range selection

原始指令

The enrollment page at lms.local (Firefox) has a dual listbox: 60 students on the left, an empty right list, with search, shift-click range select and arrow buttons. Move exactly the 18 students in fixtures/section_b.csv to the right in roster order (the right list keeps insertion order) and submit.

平台
win · Windows mac · macOS web · 任意系统(网页)
应用
Firefox · lms.local replica
交互表面
brow · 浏览器
GUI 需求 · 任务设定
G3 · 仅 GUI
GUI 需求 · 普通权限
G2 · GUI 必要
交互原语
click · 指向与点击 key · 快捷键
工作产物
WP08 · 应用 / 系统状态
条件
未声明
提案难度
3 / 5

建议验证方式

Section B enrollment equals the 18 IDs in the given order. Trusted click/shift-click log required; bulk-import endpoint disabled; DevTools disabled by policy.

建议部分得分

60% set match (F1); 30% order (Kendall tau); 10% submitted

R01.10.01.04-T14 Pick every Tuesday and Thursday of January 2027 except holidays

原始指令

In Safari, the shift-availability form at rota.local uses a multi-date calendar where a click toggles a day (no keyboard support). Select every Tuesday and Thursday in January 2027 except dates in fixtures/holidays_2027.txt, deselect the two prefilled wrong dates, and submit.

平台
mac · macOS web · 任意系统(网页)
应用
Safari · rota.local replica
交互表面
brow · 浏览器
GUI 需求 · 任务设定
G3 · 仅 GUI
GUI 需求 · 普通权限
G2 · GUI 必要
交互原语
click · 指向与点击 nav · 平移 / 缩放 / 滚动
工作产物
WP08 · 应用 / 系统状态
条件
未声明
提案难度
3 / 5

建议验证方式

Submitted date set equals the set computed from the calendar and fixture. Toggles must be trusted clicks; hidden input read-only; Web Inspector disabled via MDM profile.

建议部分得分

70% Jaccard of submitted vs expected dates; 20% prefilled wrong dates cleared; 10% submitted

R01.10.01.04-T15 Crop and straighten a team photo in a Cropper.js widget

原始指令

The profile editor at hr.local (Chrome) shows fixtures/team_photo.jpg in a free-aspect Cropper.js widget with a straighten slider. Set a 4:5 crop around the face of the person in the green jacket with about 20% margin, rotate by -3 degrees so the horizon is level, and save.

平台
mac · macOS win · Windows web · 任意系统(网页)
应用
Chrome · Cropper.js (hr.local)
交互表面
brow · 浏览器
GUI 需求 · 任务设定
G3 · 仅 GUI
GUI 需求 · 普通权限
G2 · GUI 必要
交互原语
drag · 拖放 geom · 精确几何 vis · 视觉判断
工作产物
WP08 · 应用 / 系统状态
条件
未声明
提案难度
4 / 5

建议验证方式

Saved crop vs reference rect: IoU >= 0.90 for full credit; aspect 0.80 +/- 0.01; rotation -3 +/- 0.5 deg. Crop data changes only via trusted drags; setData API unreachable with DevTools disabled.

建议部分得分

40% crop IoU; 20% aspect ratio; 30% rotation; 10% saved

R01.10.01.04-T16 Reschedule a project in a Gantt widget by dragging bars and links

原始指令

The Frappe Gantt chart at pm.local (Edge) holds 14 tasks. Apply fixtures/replan.md: drag 5 bars to new start dates, resize 3 bars to new durations, draw 4 finish-to-start links by dragging between bar endpoints, and delete one obsolete link. No successor may start before its predecessor ends.

平台
win · Windows mac · macOS web · 任意系统(网页)
应用
Edge · Frappe Gantt (pm.local)
交互表面
brow · 浏览器
GUI 需求 · 任务设定
G3 · 仅 GUI
GUI 需求 · 普通权限
G2 · GUI 必要
交互原语
drag · 拖放 geom · 精确几何 nav · 平移 / 缩放 / 滚动
工作产物
WP08 · 应用 / 系统状态
条件
long · 长流程
提案难度
5 / 5

建议验证方式

Task start/end match targets day-exact; dependency set equals target; zero constraint violations. Changes must come from trusted pointer drags (event log); date fields in the edit dialog disabled; DevTools disabled.

建议部分得分

30% bar shifts; 20% durations; 30% dependency set; 20% no constraint violations

R01.10.01.04-T17 Draw a delivery geofence polygon and depot pin on a Leaflet map

原始指令

The zone editor at zones.local (Chrome) uses Leaflet.draw over OSM tiles. Draw polygon 'Zone 7' tracing the boundary in fixtures/zone7.pdf (four named streets and a river bank) with at most 12 vertices, drop the depot marker at the address given in the PDF, and save.

平台
mac · macOS win · Windows web · 任意系统(网页)
应用
Chrome · Leaflet.draw (zones.local)
交互表面
brow · 浏览器
GUI 需求 · 任务设定
G3 · 仅 GUI
GUI 需求 · 普通权限
G2 · GUI 必要
交互原语
click · 指向与点击 nav · 平移 / 缩放 / 滚动 geom · 精确几何 vis · 视觉判断
工作产物
WP04 · 三维 / 空间模型
条件
未声明
提案难度
4 / 5

建议验证方式

Saved GeoJSON polygon IoU vs reference (full at >= 0.90); vertices <= 12; depot within 25 m of geocoded address. GeoJSON import disabled; vertices only from trusted clicks; DevTools disabled.

建议部分得分

60% polygon IoU (scaled 0.6 to 0.9); 10% vertex limit; 20% depot marker; 10% saved

R01.10.01.04-T18 Restructure a nested WordPress menu by drag-indenting items

原始指令

A WordPress Menus screen in Safari shows a 22-item primary menu. Rearrange it to match fixtures/target_menu.json: 3 top-level sections with children nested up to depth 3 (depth is set by horizontal drag offset). Do not delete and re-add items; their IDs must survive. Save the menu.

平台
mac · macOS web · 任意系统(网页)
应用
Safari · WordPress
交互表面
brow · 浏览器
GUI 需求 · 任务设定
G2 · GUI 必要
GUI 需求 · 普通权限
G2 · GUI 必要
交互原语
drag · 拖放 geom · 精确几何
工作产物
WP08 · 应用 / 系统状态
条件
未声明
提案难度
4 / 5

建议验证方式

Menu tree from wp_postmeta (_menu_item_menu_item_parent) and menu_order equals target; original item IDs kept. Change must be the Menus form POST; WP-CLI and REST credentials withheld; Web Inspector disabled.

建议部分得分

40% parent relationships; 30% sibling order; 20% item IDs preserved; 10% saved

R01.10.01.04-T19 Control pair of T05: reorder the backlog with a Playwright script

原始指令

Control pair of T05. Same plan.local backlog and fixtures/priorities.txt, but you only have a terminal with Python Playwright and headless Chromium. Write reorder.py that drags items into the target order with real mouse events (page.mouse), clicks Save Order and exits 0. Do not call the backend directly.

平台
lin · Linux
应用
Playwright · Chromium
交互表面
term · 终端
GUI 需求 · 任务设定
CL · CLI 足够
GUI 需求 · 普通权限
G2 · GUI 必要
交互原语
code · 代码与命令
工作产物
WP06 · 代码 / 工程
条件
未声明
提案难度
3 / 5

建议验证方式

Grader runs reorder.py on a fresh seed with a different shuffle; backend order must equal fixture. Proxy rejects POSTs not preceded by page drag telemetry. Score compared with T05 for the GUI/CLI gap.

建议部分得分

50% script runs on the new shuffle; 40% exact order; 10% saved through the page

R01.10.01.04-T20 Complete a travel-request form built from nine widget types

原始指令

Fill the 4-step travel request at corp.local (Chrome) from fixtures/request.json: range date picker, two airport autocompletes, a per-diem dual slider, grouped cost-center multi-select, time picker, cascading country/city selects, drag-ordered list of 4 meetings, a drop zone for agenda.pdf and a slide-to-confirm control. Submit; values must survive back/next navigation.

平台
mac · macOS win · Windows web · 任意系统(网页)
应用
Chrome · corp.local replica
交互表面
brow · 浏览器
GUI 需求 · 任务设定
G3 · 仅 GUI
GUI 需求 · 普通权限
G2 · GUI 必要
交互原语
drag · 拖放 type · 文字与表单 click · 指向与点击 geom · 精确几何
工作产物
WP02 · 记录 / 表格
条件
long · 长流程
提案难度
5 / 5

建议验证方式

Submitted record compared field by field to fixture (exact; slider +/- 0); uploaded file SHA-256 matches. Each widget logs trusted events; a programmatically set field scores 0. DevTools disabled; draft API hidden.

建议部分得分

90% nine widgets correct, equal shares; 10% submission with no field reset after step navigation

已有研究映射

以下沿用研究目录中的基准与小类映射,不是逐题标注、执行记录或产品背书。

同工作流族的小类