ComponentBench / tasks /v1 /file_upload_button.yaml
TianchenGuan's picture
Initial ComponentBench data release: tasks v1+v2, cleaned human traces, difficulty audit, ontology metadata
360df42 verified
- id: file_upload_button-antd-T01
name: Upload a resume PDF (single file)
canonical_type: file_upload_button
implementation_source: antd
implementation_variant: Upload (default button)
implementation_component: 'AntD: Upload'
task_template: file_upload
secondary_template: null
browsergym_goal: Upload the file "resume.pdf" using the Resume uploader. The task will finish automatically when done.
ui_copy: 'Resume uploader: Upload "resume.pdf".'
setup_description: 'The page shows a single isolated card titled "Job application – Resume" centered in the viewport. Inside
the card there is one Ant Design Upload component rendered as a prominent primary button labeled "Select file" with a
small hint text "PDF only" under it. The Upload uses the default (text) listType and shows the upload list below the button
once a file is chosen. The component is configured for a single file (multiple=false) and auto-uploads immediately to
a mocked endpoint; progress briefly shows as "uploading" then becomes "done". No other interactive elements are present
(no extra form fields, no extra uploaders). Initial state: the upload list is empty and no errors are shown.'
scene_context:
theme: light
spacing: comfortable
layout: isolated_card
placement: center
scale: default
instances: 1
guidance: text
clutter: none
difficulty:
difficulty_bucket: easy
tier: L0
axes_ratings:
precision_requirement: 2
target_acquisition: 1
density_choice_interference: 1
depth_layering: 1
feedback_dynamics: 2
semantic_observability: 1
disambiguation_load: 1
justification: Single, clearly labeled uploader with immediate feedback in the visible file list; only one file is required.
success_trigger:
human_readable:
- The Resume uploader's file list contains exactly one file named "resume.pdf".
- The file's upload status is "done" (not "uploading" or "error").
- No other file is present in the uploader.
canonical_predicate:
predicate_type: equals
target_state:
files:
- resume.pdf
count: 1
status:
resume.pdf: done
tolerance: null
require_confirm: false
confirm_control: null
require_correct_instance: false
target_instance_label_or_id: null
terminal_condition: task ends when predicate holds
negative_cases:
- A different file is uploaded (wrong filename).
- More than one file is present in the upload list.
- '"resume.pdf" is present but still in an "uploading"/"error" state.'
- The file is uploaded in some other (non-target) uploader (there is only one on this page, but guard against accidental
global matching).
expected_interaction_path:
- Click the "Select file" upload button
- Choose "resume.pdf" in the file picker
- Wait until the upload list shows "resume.pdf" with status done
notes: 'Checker hint: read AntD Upload fileList and status from the component instance data-testid="uploader-resume".'
- id: file_upload_button-antd-T02
name: Upload profile photo to picture-card uploader (mixed guidance)
canonical_type: file_upload_button
implementation_source: antd
implementation_variant: Upload (picture-card, maxCount=1)
implementation_component: 'AntD: Upload (listType="picture-card")'
task_template: file_upload
secondary_template: null
browsergym_goal: Upload the file "profile_photo.png" using the Profile photo uploader. The task will finish automatically
when done.
ui_copy: 'Profile photo: Upload "profile_photo.png" (PNG/JPG).'
setup_description: 'A centered isolated card titled "Account – Profile photo" contains an Ant Design Upload configured as
a picture-card (thumbnail grid). The uploader shows a single empty picture-card slot with a plus icon and the text "Upload";
maxCount is 1 so only one image can be stored. Accepted types are limited to PNG/JPEG, and the helper text below the component
says "Images only". To the right of the uploader, a non-interactive Reference panel shows the expected image thumbnail
and the filename text "profile_photo.png" (this provides mixed text+visual guidance). After selection, the chosen image
appears as a thumbnail tile inside the Upload component with small overlay icons for preview and remove. Initial state:
no file is selected, and the upload list grid is empty.'
scene_context:
theme: light
spacing: comfortable
layout: isolated_card
placement: center
scale: default
instances: 1
guidance: mixed
clutter: none
difficulty:
difficulty_bucket: easy
tier: L0
axes_ratings:
precision_requirement: 2
target_acquisition: 2
density_choice_interference: 1
depth_layering: 1
feedback_dynamics: 2
semantic_observability: 1
disambiguation_load: 1
justification: Still a single uploader with clear label; picture-card icons are slightly smaller, but the target file
is explicitly named and visually referenced.
success_trigger:
human_readable:
- The Profile photo uploader contains exactly one file named "profile_photo.png".
- The thumbnail tile is present and the file status is "done".
- No second image is present (maxCount=1).
canonical_predicate:
predicate_type: equals
target_state:
files:
- profile_photo.png
count: 1
status:
profile_photo.png: done
tolerance: null
require_confirm: false
confirm_control: null
require_correct_instance: false
target_instance_label_or_id: null
terminal_condition: task ends when predicate holds
negative_cases:
- Uploading a different image file (wrong filename).
- The file appears but is in an "error" state (e.g., rejected type).
- More than one file is present (should be impossible with maxCount=1, but do not count success if it occurs).
- Only opening the preview without actually selecting/uploading the file.
expected_interaction_path:
- Click the picture-card "Upload" tile
- Choose "profile_photo.png" in the file picker
- Wait for the thumbnail tile to show as uploaded (done)
notes: 'Checker hint: use data-testid="uploader-profile-photo" and read Upload fileList names + status.'
- id: file_upload_button-antd-T03
name: Remove an already uploaded file (no confirmation)
canonical_type: file_upload_button
implementation_source: antd
implementation_variant: Upload (default list with remove icon)
implementation_component: 'AntD: Upload'
task_template: file_manage
secondary_template: null
browsergym_goal: Remove the file "draft_cover_letter.docx" from the Attachment uploader so that no files remain. The task
will finish automatically when done.
ui_copy: 'Attachment uploader: remove "draft_cover_letter.docx".'
setup_description: The page contains one centered isolated card titled "Attachments" with a single Ant Design Upload component
in the default (text) list style. The upload list is visible on page load and already contains one completed item named
"draft_cover_letter.docx" with status "done". Each list item has a small remove (×) icon on the right. There is no confirmation
dialog for removal; clicking the remove icon immediately removes the item from the list. No other uploaders or controls
are present.
scene_context:
theme: light
spacing: comfortable
layout: isolated_card
placement: center
scale: default
instances: 1
guidance: text
clutter: none
difficulty:
difficulty_bucket: easy
tier: L0
axes_ratings:
precision_requirement: 2
target_acquisition: 2
density_choice_interference: 1
depth_layering: 1
feedback_dynamics: 1
semantic_observability: 1
disambiguation_load: 1
justification: Single file to remove with a direct remove affordance and immediate visible state change.
success_trigger:
human_readable:
- The Attachment uploader's file list is empty (0 files).
- No hidden/stale file remains selected in the underlying input.
canonical_predicate:
predicate_type: equals
target_state:
files: []
count: 0
tolerance: null
require_confirm: false
confirm_control: null
require_correct_instance: false
target_instance_label_or_id: null
terminal_condition: task ends when predicate holds
negative_cases:
- The file remains in the list (removal not completed).
- A different file is removed/affected (not applicable here since only one file exists, but guard against global matching).
- The file is marked as removed but still present in the list UI.
- Removal is attempted but canceled (there is no confirm in this task).
expected_interaction_path:
- Locate "draft_cover_letter.docx" in the upload list
- Click the remove (×) icon on that list item
- Verify the list becomes empty
notes: 'Checker hint: treat success as fileList.length==0 for data-testid="uploader-attachment".'
- id: file_upload_button-antd-T04
name: Upload to the correct uploader when two are present
canonical_type: file_upload_button
implementation_source: antd
implementation_variant: Upload (two instances)
implementation_component: 'AntD: Upload'
task_template: file_upload
secondary_template: null
browsergym_goal: Upload the file "resume.pdf" using the uploader labeled "Resume" (not the "Profile photo" uploader). The
task will finish automatically when done.
ui_copy: 'Two uploaders: Profile photo and Resume. Upload "resume.pdf" to Resume.'
setup_description: A single isolated card titled "Candidate profile" is centered on the page. Inside the card there are
two clearly separated Ant Design Upload components of the same canonical type. The first section is labeled "Profile photo"
and uses a picture-card style with image-only accept rules (PNG/JPG) and maxCount=1; it starts empty. The second section
is labeled "Resume" and uses the default text list style and accepts PDF; it also starts empty. Both sections have their
own "Select file" button and their own independent upload list area directly below. No other controls are on the page.
scene_context:
theme: light
spacing: comfortable
layout: isolated_card
placement: center
scale: default
instances: 2
guidance: text
clutter: none
difficulty:
difficulty_bucket: easy
tier: L1
axes_ratings:
precision_requirement: 2
target_acquisition: 2
density_choice_interference: 1
depth_layering: 1
feedback_dynamics: 2
semantic_observability: 1
disambiguation_load: 2
justification: Requires choosing the correct instance among two labeled uploaders, but labels are clear and the target
file is explicit.
success_trigger:
human_readable:
- The uploader labeled "Resume" contains exactly one file named "resume.pdf" with status "done".
- The uploader labeled "Profile photo" remains empty (0 files).
canonical_predicate:
predicate_type: equals
target_state:
instance: Resume
files:
- resume.pdf
count: 1
status:
resume.pdf: done
other_instances_unchanged:
Profile photo:
count: 0
tolerance: null
require_confirm: false
confirm_control: null
require_correct_instance: true
target_instance_label_or_id: Resume
terminal_condition: task ends when predicate holds
negative_cases:
- '"resume.pdf" is uploaded to the Profile photo uploader (wrong instance).'
- The Resume uploader contains a different file or multiple files.
- '"resume.pdf" is present but still uploading or in error.'
- Both uploaders are modified (e.g., a file is also added to Profile photo).
expected_interaction_path:
- Locate the section labeled "Resume"
- Click its "Select file" button
- Choose "resume.pdf" in the file picker
- Wait until the Resume list shows status done
notes: 'Instrumentation: assign data-testid="uploader-profile-photo" and data-testid="uploader-resume"; checker must bind
to the Resume instance.'
- id: file_upload_button-antd-T05
name: Confirm removal in Popconfirm (dark form section)
canonical_type: file_upload_button
implementation_source: antd
implementation_variant: Upload + Popconfirm on remove
implementation_component: 'AntD: Upload + Popconfirm'
task_template: confirm_cancel
secondary_template: file_manage
browsergym_goal: In the uploader labeled "Receipts", remove the file "invoice_2025-11.pdf" and confirm the removal by clicking
"Remove" in the confirmation pop-up. The task will finish automatically when done.
ui_copy: 'Receipts uploader: remove "invoice_2025-11.pdf" and confirm "Remove".'
setup_description: 'The page is a small "Expense report" form section in dark theme. Several non-required inputs (Amount,
Category, Notes) appear above, and the target uploader appears below under a label "Receipts". The Ant Design Upload component
uses the default text list, and it loads with one existing completed file: "invoice_2025-11.pdf" (status done). Clicking
the remove (×) icon does not immediately remove the file; instead it opens an Ant Design Popconfirm anchored near the
list item. The pop-up contains the message "Remove this file?" and two buttons: "Cancel" and "Remove". Only after clicking
"Remove" is the file actually removed from the upload list.'
scene_context:
theme: dark
spacing: comfortable
layout: form_section
placement: center
scale: default
instances: 1
guidance: text
clutter: low
difficulty:
difficulty_bucket: mid
tier: L1
axes_ratings:
precision_requirement: 2
target_acquisition: 3
density_choice_interference: 1
depth_layering: 2
feedback_dynamics: 2
semantic_observability: 2
disambiguation_load: 2
justification: Requires hitting a small remove icon and then correctly interacting with a confirmation pop-up in a slightly
cluttered form context.
success_trigger:
human_readable:
- In the "Receipts" uploader, the file list becomes empty (0 files).
- The confirmation pop-up has been accepted via the "Remove" button (not canceled).
canonical_predicate:
predicate_type: equals
target_state:
instance: Receipts
files: []
count: 0
confirmation:
type: popconfirm
accepted: true
tolerance: null
require_confirm: true
confirm_control: Remove
require_correct_instance: true
target_instance_label_or_id: Receipts
terminal_condition: task ends when predicate holds
negative_cases:
- Clicking remove but then pressing "Cancel" (file remains).
- The file is still present after closing the pop-up.
- Removing a different file (if additional files are uploaded).
- Removing the file from another uploader (there is only one uploader in this task, but enforce instance binding to "Receipts").
expected_interaction_path:
- Find "invoice_2025-11.pdf" in the Receipts list
- Click its remove (×) icon
- In the popconfirm, click "Remove"
- Verify the list is empty
notes: 'Checker hint: instrument popconfirm result (accepted/canceled) or assert list empty after a confirm click. data-testid="uploader-receipts".'
- id: file_upload_button-antd-T06
name: Upload the file that matches a reference card (visual guidance)
canonical_type: file_upload_button
implementation_source: antd
implementation_variant: Upload (default) + reference panel
implementation_component: 'AntD: Upload'
task_template: match_reference
secondary_template: null
browsergym_goal: Upload the same file that is shown in the Reference card using the uploader labeled "Attachment". The task
will finish automatically when done.
ui_copy: 'Attachment uploader: upload the file shown in the Reference card.'
setup_description: The main content is anchored near the top-left of the viewport (not centered). An isolated card titled
"Attach a file" contains the target Ant Design Upload component labeled "Attachment" with a "Select file" button and an
empty upload list. Next to the uploader, a non-interactive Reference card shows the target file (a file icon plus the
filename text). The Reference card is the only place where the target filename is provided; the instruction intentionally
refers to the reference instead of naming the file. The Upload component is single-file (multiple=false) and auto-uploads;
after selection, the file appears in the list with status done.
scene_context:
theme: light
spacing: comfortable
layout: isolated_card
placement: top_left
scale: default
instances: 1
guidance: visual
clutter: none
difficulty:
difficulty_bucket: mid
tier: L1
axes_ratings:
precision_requirement: 2
target_acquisition: 2
density_choice_interference: 2
depth_layering: 1
feedback_dynamics: 2
semantic_observability: 2
disambiguation_load: 1
justification: Requires extracting the target filename from an on-screen reference and then completing a standard upload
with visible list feedback.
success_trigger:
human_readable:
- The "Attachment" uploader contains exactly one uploaded file.
- The uploaded file's name matches the filename displayed in the Reference card.
- The uploaded file's status is "done".
canonical_predicate:
predicate_type: matches_reference
target_state:
instance: Attachment
reference_slot: Reference card
match_field: file_name
count: 1
status: done
tolerance: null
require_confirm: false
confirm_control: null
require_correct_instance: true
target_instance_label_or_id: Attachment
terminal_condition: task ends when predicate holds
negative_cases:
- Uploading any file that does not match the Reference card filename.
- Uploading more than one file.
- Uploading the correct file but leaving it in "uploading" or "error" state.
- Uploading into the wrong component instance (should only bind to the uploader labeled "Attachment").
expected_interaction_path:
- Read the filename shown in the Reference card
- Click "Select file" in the Attachment uploader
- Choose that file in the file picker
- Wait until the list shows status done
notes: 'Checker hint: store reference filename in data-testid="reference-file-name" and compare to Upload fileList[0].name
for data-testid="uploader-attachment".'
- id: file_upload_button-antd-T07
name: Drag-and-drop upload with Upload.Dragger (compact spacing)
canonical_type: file_upload_button
implementation_source: antd
implementation_variant: Upload.Dragger
implementation_component: 'AntD: Upload.Dragger'
task_template: drag_operation
secondary_template: file_upload
browsergym_goal: Drag and drop the file "diagram.svg" into the drag-and-drop uploader. The task will finish automatically
when done.
ui_copy: 'Drag-and-drop uploader: add "diagram.svg" by dragging it into the drop area.'
setup_description: A centered isolated card titled "Upload asset" contains an Ant Design Upload.Dragger component (large
dashed drop area). The page is in compact spacing mode, so padding around the drop area and surrounding text is reduced.
Inside the drop area, the helper text reads "Click or drag file to this area to upload" and there is no separate button.
A small, non-interactive "Test files" note below the card lists available files including "diagram.svg". On drop, the
file is added to a file list rendered below the drop area and quickly transitions to status done.
scene_context:
theme: light
spacing: compact
layout: isolated_card
placement: center
scale: default
instances: 1
guidance: text
clutter: none
difficulty:
difficulty_bucket: mid
tier: L1
axes_ratings:
precision_requirement: 2
target_acquisition: 3
density_choice_interference: 2
depth_layering: 1
feedback_dynamics: 2
semantic_observability: 2
disambiguation_load: 1
justification: Requires performing a drag-and-drop interaction on a large target area; compact spacing slightly increases
interaction tightness.
success_trigger:
human_readable:
- The drag-and-drop uploader contains exactly one file named "diagram.svg".
- The file's status is "done".
- The file was provided via a drop interaction (drop event observed).
canonical_predicate:
predicate_type: equals
target_state:
files:
- diagram.svg
count: 1
status:
diagram.svg: done
interaction_method: drop
tolerance: null
require_confirm: false
confirm_control: null
require_correct_instance: false
target_instance_label_or_id: null
terminal_condition: task ends when predicate holds
negative_cases:
- The file list remains empty (drop not completed).
- A different file is uploaded.
- '"diagram.svg" is present but in "uploading" or "error" state.'
- The file was added via click-to-select instead of drop (if the task run enforces drop-only).
expected_interaction_path:
- Drag the file "diagram.svg" onto the dashed drop area
- Release to drop
- Wait until the file appears in the list with status done
notes: 'Checker hint: instrument Upload onDrop handler to set data-last-input="drop" for this component; validate it alongside
fileList.'
- id: file_upload_button-antd-T08
name: Manual upload with Start upload in a settings panel (3 instances)
canonical_type: file_upload_button
implementation_source: antd
implementation_variant: Upload (beforeUpload=false manual start)
implementation_component: 'AntD: Upload + Button("Start upload")'
task_template: file_upload
secondary_template: confirm_cancel
browsergym_goal: In the settings panel, upload the file "report_final.pdf" using the uploader labeled "Resume (PDF)" and
then click "Start upload" for that uploader. The task will finish automatically when done.
ui_copy: 'Resume (PDF): add "report_final.pdf" and click Start upload.'
setup_description: 'The page uses a settings-panel layout titled "Profile settings" with multiple subsections and medium
clutter (toggles and text inputs are visible but irrelevant). In the "Documents" subsection there are three Ant Design
Upload instances stacked vertically with headings: "Profile photo", "Resume (PDF)", and "Cover letter". "Profile photo"
is a picture-card uploader and already contains one image file (pre-filled) to act as a distractor. "Resume (PDF)" is
the target uploader and starts empty; it is configured for manual upload: selecting a file adds it to the list in a "ready"
state and does not upload automatically. Directly below the Resume uploader''s file list is a small primary button labeled
"Start upload" that must be clicked to complete the upload. "Cover letter" is another text-style uploader with an existing
file in its list. Only the Resume uploader''s file list determines success for this task.'
scene_context:
theme: light
spacing: comfortable
layout: settings_panel
placement: center
scale: default
instances: 3
guidance: text
clutter: medium
difficulty:
difficulty_bucket: hard
tier: L2
axes_ratings:
precision_requirement: 3
target_acquisition: 3
density_choice_interference: 3
depth_layering: 2
feedback_dynamics: 3
semantic_observability: 2
disambiguation_load: 4
justification: Multiple similar uploaders require correct disambiguation, and the manual two-step flow (select Start
upload) increases layering and feedback complexity.
success_trigger:
human_readable:
- The uploader labeled "Resume (PDF)" contains exactly one file named "report_final.pdf".
- '"report_final.pdf" has completed uploading (status "done"), which only happens after clicking "Start upload".'
- Other uploader instances are not modified (their existing files remain unchanged).
canonical_predicate:
predicate_type: equals
target_state:
instance: Resume (PDF)
files:
- report_final.pdf
count: 1
status:
report_final.pdf: done
confirm_flow:
confirm_control: Start upload
activated: true
other_instances_unchanged:
Profile photo:
count: 1
Cover letter:
count: 1
tolerance: null
require_confirm: true
confirm_control: Start upload
require_correct_instance: true
target_instance_label_or_id: Resume (PDF)
terminal_condition: task ends when predicate holds
negative_cases:
- '"report_final.pdf" is selected but remains in a pre-upload/ready state (Start upload not clicked).'
- Uploading the file into "Profile photo" or "Cover letter" instead of "Resume (PDF)".
- Uploading the wrong file, or uploading multiple files to the Resume uploader.
- Modifying other uploader instances (e.g., removing their existing files).
expected_interaction_path:
- Locate the "Resume (PDF)" uploader in the Documents section
- Click its "Select file" button and choose "report_final.pdf"
- Click the Resume uploader's "Start upload" button
- Wait until the list shows status done
notes: 'Checker hint: store per-instance data-testid (uploader-profile-photo, uploader-resume, uploader-cover-letter) and
instrument Start upload click per instance.'
- id: file_upload_button-antd-T09
name: Replace an existing logo image (dark theme, 2 instances)
canonical_type: file_upload_button
implementation_source: antd
implementation_variant: Upload (maxCount=1 replace)
implementation_component: 'AntD: Upload (maxCount=1, picture-card)'
task_template: file_manage
secondary_template: file_upload
browsergym_goal: In the uploader labeled "Company logo", replace the current file "logo_old.png" with "logo_new.png" so
that only "logo_new.png" remains. The task will finish automatically when done.
ui_copy: 'Company logo: replace logo_old.png with logo_new.png.'
setup_description: 'A dark-themed centered card titled "Brand assets" contains two Ant Design Upload components side-by-side.
Left: "Company logo" uses picture-card thumbnails with maxCount=1 and already has one uploaded image tile labeled "logo_old.png".
Right: "Team photo" is another picture-card uploader that starts empty and is a distractor. In the Company logo tile,
small overlay icons for preview and remove appear on hover. Selecting a new file while maxCount=1 is set replaces the
existing file; alternatively, the user may remove the old file and then upload the new one.'
scene_context:
theme: dark
spacing: comfortable
layout: isolated_card
placement: center
scale: default
instances: 2
guidance: text
clutter: none
difficulty:
difficulty_bucket: hard
tier: L2
axes_ratings:
precision_requirement: 3
target_acquisition: 3
density_choice_interference: 2
depth_layering: 2
feedback_dynamics: 3
semantic_observability: 2
disambiguation_load: 3
justification: Requires correctly targeting the "Company logo" instance and ensuring the final list contains only the
new file, with small picture-card affordances in dark theme.
success_trigger:
human_readable:
- The uploader labeled "Company logo" contains exactly one file named "logo_new.png" with status "done".
- '"logo_old.png" is not present in the Company logo uploader.'
- The uploader labeled "Team photo" remains empty.
canonical_predicate:
predicate_type: equals
target_state:
instance: Company logo
files:
- logo_new.png
count: 1
status:
logo_new.png: done
not_present:
- logo_old.png
other_instances_unchanged:
Team photo:
count: 0
tolerance: null
require_confirm: false
confirm_control: null
require_correct_instance: true
target_instance_label_or_id: Company logo
terminal_condition: task ends when predicate holds
negative_cases:
- Both files are present (logo_old.png not removed).
- The wrong file is uploaded (not logo_new.png).
- The file is uploaded to the Team photo uploader (wrong instance).
- '"logo_new.png" is present but failed or still uploading.'
expected_interaction_path:
- Locate the "Company logo" uploader
- Either upload "logo_new.png" directly to replace, or remove "logo_old.png" then upload "logo_new.png"
- Verify only logo_new.png remains and is done
notes: 'Checker hint: enforce instance scoping via data-testid="uploader-company-logo"; validate that fileList names exactly
match.'
- id: file_upload_button-antd-T10
name: Clear all uploaded files from a long list (dashboard)
canonical_type: file_upload_button
implementation_source: antd
implementation_variant: Upload (multiple) + Clear all action
implementation_component: 'AntD: Upload (multiple=true) + custom Clear all control'
task_template: clear_reset
secondary_template: confirm_cancel
browsergym_goal: Clear all files from the uploader labeled "Attachments" and confirm the clear action if prompted. The task
will finish automatically when done.
ui_copy: 'Attachments uploader: clear all files (confirm if asked).'
setup_description: 'The page is a dashboard layout with high clutter: a left navigation menu, a top header, and several
statistic cards. In the main area, a card titled "Attachments" contains the target Ant Design Upload component in the
default text-list style with multiple files enabled. On load, the upload list already contains a long set of files (about
8–12 items) and the list area is vertically scrollable within the card. Above the list there is a small text link/button
labeled "Clear all" (part of the uploader''s header controls). Clicking "Clear all" opens a confirmation pop-up asking
to confirm clearing all files, with buttons "Cancel" and "Clear". After confirming, the list should become empty.'
scene_context:
theme: light
spacing: comfortable
layout: dashboard
placement: center
scale: default
instances: 1
guidance: text
clutter: high
difficulty:
difficulty_bucket: hard
tier: L3
axes_ratings:
precision_requirement: 2
target_acquisition: 3
density_choice_interference: 3
depth_layering: 2
feedback_dynamics: 3
semantic_observability: 2
disambiguation_load: 2
justification: A long, scrollable file list in a cluttered dashboard plus a confirmation step increases target acquisition
and recovery complexity.
success_trigger:
human_readable:
- The "Attachments" uploader shows 0 files in its file list (completely cleared).
- If a confirmation pop-up appears, it has been accepted (not canceled).
canonical_predicate:
predicate_type: equals
target_state:
instance: Attachments
files: []
count: 0
confirmation:
type: popconfirm
accepted: true
tolerance: null
require_confirm: true
confirm_control: Clear
require_correct_instance: true
target_instance_label_or_id: Attachments
terminal_condition: task ends when predicate holds
negative_cases:
- Some files remain in the list after the action (partial clear).
- The clear action is canceled in the confirmation dialog.
- Files are removed one-by-one but at least one remains (still not success).
- Clearing happens in a different uploader (guard against accidental matching if the page template is reused).
expected_interaction_path:
- Locate the "Attachments" uploader card within the dashboard
- Click the "Clear all" control
- If prompted, click "Clear" to confirm
- Verify the upload list becomes empty
notes: 'Checker hint: assert fileList length==0 for data-testid="uploader-attachments"; optionally assert a confirm click
event was recorded.'
- id: file_upload_button-mui-T11
name: Upload a resume PDF with MUI Button-as-label
canonical_type: file_upload_button
implementation_source: mui
implementation_variant: 'composite: Button(label)+hidden input'
implementation_component: 'MUI: Button (component="label") + visually hidden <input type="file">'
task_template: file_upload
secondary_template: null
browsergym_goal: Upload the file "resume.pdf" using the button labeled "Upload resume". The task will finish automatically
when done.
ui_copy: 'Upload resume: choose "resume.pdf".'
setup_description: 'A centered isolated card titled "Resume upload" contains a single MUI contained Button labeled "Upload
resume". The button is implemented as a label (component="label") wrapping a visually hidden <input type="file"> (standard
MUI file upload pattern). The input accepts a single PDF file (accept="application/pdf") and does not allow multiple selection.
After a file is selected, the chosen filename is displayed directly below the button in a one-line helper text (Typography)
that reads "Selected: <filename>". There are no other interactive elements on the page; initial state shows "Selected:
none".'
scene_context:
theme: light
spacing: comfortable
layout: isolated_card
placement: center
scale: default
instances: 1
guidance: text
clutter: none
difficulty:
difficulty_bucket: easy
tier: L0
axes_ratings:
precision_requirement: 2
target_acquisition: 1
density_choice_interference: 1
depth_layering: 1
feedback_dynamics: 1
semantic_observability: 1
disambiguation_load: 1
justification: Single large button with immediate visible filename feedback; straightforward single-file selection.
success_trigger:
human_readable:
- The hidden file input associated with "Upload resume" contains exactly one file named "resume.pdf".
- 'The on-page helper text reflects "Selected: resume.pdf".'
canonical_predicate:
predicate_type: equals
target_state:
files:
- resume.pdf
count: 1
tolerance: null
require_confirm: false
confirm_control: null
require_correct_instance: false
target_instance_label_or_id: null
terminal_condition: task ends when predicate holds
negative_cases:
- A different file is selected.
- 'No file is selected (still "Selected: none").'
- Multiple files are selected (should be disallowed).
- The filename text updates but the underlying input has no file (stale UI).
expected_interaction_path:
- Click the "Upload resume" button
- Choose "resume.pdf" in the file picker
- Verify the helper text shows the selected filename
notes: 'Checker hint: query the input[type=file] linked to data-testid="uploader-resume" and validate input.files[0].name.'
- id: file_upload_button-mui-T12
name: Upload a profile picture via small IconButton (corner placement)
canonical_type: file_upload_button
implementation_source: mui
implementation_variant: 'composite: IconButton(label)+hidden input'
implementation_component: 'MUI: IconButton + visually hidden <input type="file">'
task_template: file_upload
secondary_template: null
browsergym_goal: Upload the image file "profile_photo.png" using the control labeled "Change photo". The task will finish
automatically when done.
ui_copy: 'Change photo: upload "profile_photo.png".'
setup_description: The upload UI is anchored near the bottom-left of the viewport. A small card titled "Profile picture"
shows a placeholder avatar and a compact MUI IconButton (camera icon) labeled with aria-label "Change photo". The IconButton
is implemented as a label wrapping a hidden input[type=file] that accepts images (PNG/JPG). When a file is chosen, the
avatar preview updates and a caption below the avatar shows the selected filename. No other uploaders are present.
scene_context:
theme: light
spacing: comfortable
layout: isolated_card
placement: bottom_left
scale: default
instances: 1
guidance: text
clutter: none
difficulty:
difficulty_bucket: easy
tier: L1
axes_ratings:
precision_requirement: 2
target_acquisition: 3
density_choice_interference: 1
depth_layering: 1
feedback_dynamics: 2
semantic_observability: 2
disambiguation_load: 1
justification: Small IconButton target increases acquisition difficulty, but only one uploader and a single explicit file
are required.
success_trigger:
human_readable:
- The "Change photo" file input contains exactly one file named "profile_photo.png".
- The preview caption (or avatar alt text) indicates the selected file is profile_photo.png.
canonical_predicate:
predicate_type: equals
target_state:
files:
- profile_photo.png
count: 1
tolerance: null
require_confirm: false
confirm_control: null
require_correct_instance: false
target_instance_label_or_id: null
terminal_condition: task ends when predicate holds
negative_cases:
- A different image file is selected.
- No file is selected.
- The avatar preview changes but the underlying input file is not profile_photo.png.
- Selecting a non-image file that is rejected.
expected_interaction_path:
- Click the small camera IconButton labeled "Change photo"
- Choose "profile_photo.png" in the file picker
- Verify the caption/preview reflects the selected file
notes: 'Instrumentation: ensure IconButton has aria-label="Change photo" and its associated input has data-testid="uploader-profile-photo".'
- id: file_upload_button-mui-T13
name: Clear a selected file using Chip delete (reset)
canonical_type: file_upload_button
implementation_source: mui
implementation_variant: 'composite: Button+file chip list'
implementation_component: 'MUI: Button(label)+hidden input + Chip'
task_template: clear_reset
secondary_template: file_manage
browsergym_goal: Clear the current selection in the uploader labeled "Attachment" so that no file is selected. The task
will finish automatically when done.
ui_copy: 'Attachment: clear the selected file (currently notes.txt).'
setup_description: A centered isolated card titled "Attachment" contains a MUI Button labeled "Choose file" (Button-as-label
with hidden input[type=file]). Below the button, the currently selected file is displayed as a MUI Chip reading "notes.txt"
with a small delete (×) icon on the right. The page loads with notes.txt already selected (the hidden input has that file
in its FileList). Clicking the chip delete icon clears the selection and the UI switches back to a placeholder text "No
file selected".
scene_context:
theme: light
spacing: comfortable
layout: isolated_card
placement: center
scale: default
instances: 1
guidance: text
clutter: none
difficulty:
difficulty_bucket: easy
tier: L0
axes_ratings:
precision_requirement: 2
target_acquisition: 2
density_choice_interference: 1
depth_layering: 1
feedback_dynamics: 1
semantic_observability: 1
disambiguation_load: 1
justification: Single visible chip delete affordance clears the selection with immediate feedback and no extra steps.
success_trigger:
human_readable:
- The Attachment uploader has 0 selected files (input.files.length == 0).
- The on-page state indicates "No file selected" (chip is removed).
canonical_predicate:
predicate_type: equals
target_state:
files: []
count: 0
tolerance: null
require_confirm: false
confirm_control: null
require_correct_instance: false
target_instance_label_or_id: null
terminal_condition: task ends when predicate holds
negative_cases:
- The chip remains visible and the input still contains notes.txt.
- A different file remains selected instead of clearing all files.
- Clearing is attempted but the UI text changes without clearing the underlying input (stale value).
expected_interaction_path:
- Find the chip showing the selected file name
- Click the chip's delete (×) icon
- Verify the chip disappears and selection is empty
notes: 'Checker hint: validate both the rendered chip absence and the input.files length for data-testid="uploader-attachment".'
- id: file_upload_button-mui-T14
name: Upload to the correct field in a form (2 instances)
canonical_type: file_upload_button
implementation_source: mui
implementation_variant: 'composite: two Button-as-label uploaders'
implementation_component: 'MUI: Button(label)+hidden input (two instances)'
task_template: file_upload
secondary_template: null
browsergym_goal: In the form, upload the file "invoice_2025-11.pdf" using the uploader labeled "Proof of address" (not "Passport
scan"). The task will finish automatically when done.
ui_copy: 'Proof of address: upload invoice_2025-11.pdf.'
setup_description: 'The page is a short identity verification form section (low clutter) with two upload fields placed one
under another. The first uploader is labeled "Passport scan" and the second is labeled "Proof of address"; both are MUI
Buttons implemented as labels wrapping hidden file inputs. Each field shows its own selected filename line below the button
(e.g., "Selected: none"). Both uploaders start empty. Other form fields (name, date of birth) are present above but are
irrelevant to success.'
scene_context:
theme: light
spacing: comfortable
layout: form_section
placement: center
scale: default
instances: 2
guidance: text
clutter: low
difficulty:
difficulty_bucket: mid
tier: L1
axes_ratings:
precision_requirement: 2
target_acquisition: 2
density_choice_interference: 1
depth_layering: 1
feedback_dynamics: 1
semantic_observability: 1
disambiguation_load: 3
justification: Disambiguation between two similar upload controls in a form is the main challenge; feedback is clear once
the correct field is used.
success_trigger:
human_readable:
- The uploader labeled "Proof of address" has exactly one selected file named "invoice_2025-11.pdf".
- The uploader labeled "Passport scan" remains empty.
canonical_predicate:
predicate_type: equals
target_state:
instance: Proof of address
files:
- invoice_2025-11.pdf
count: 1
other_instances_unchanged:
Passport scan:
count: 0
tolerance: null
require_confirm: false
confirm_control: null
require_correct_instance: true
target_instance_label_or_id: Proof of address
terminal_condition: task ends when predicate holds
negative_cases:
- '"invoice_2025-11.pdf" is selected in the Passport scan uploader (wrong instance).'
- The Proof of address uploader has a different file or multiple files.
- Both uploaders are filled (should only fill Proof of address).
expected_interaction_path:
- Locate the "Proof of address" uploader section
- Click its "Choose file"/"Upload" button
- Select "invoice_2025-11.pdf"
- Confirm the selected filename appears under the Proof of address field
notes: 'Instrumentation: data-testid="uploader-passport" and data-testid="uploader-proof-address" for instance-scoped checking.'
- id: file_upload_button-mui-T15
name: Confirm removal in a MUI Dialog
canonical_type: file_upload_button
implementation_source: mui
implementation_variant: 'composite: upload list + confirm dialog on remove'
implementation_component: 'MUI: Button(label)+hidden input + List + Dialog'
task_template: confirm_cancel
secondary_template: file_manage
browsergym_goal: Remove the file "data_sample.csv" from the uploader labeled "Data attachment" and confirm by clicking "Remove"
in the dialog. The task will finish automatically when done.
ui_copy: 'Data attachment: remove data_sample.csv (confirm Remove).'
setup_description: 'A centered isolated card titled "Data attachment" contains a MUI upload control (Button-as-label with
hidden input) plus a small list area that displays selected/uploaded files. On page load, the list contains exactly one
item: "data_sample.csv" with a trailing trash icon button labeled "Remove file". Clicking the trash icon opens a MUI Dialog
titled "Remove file?" with two buttons: "Cancel" and "Remove". The file is only removed from the list after clicking the
"Remove" button in the dialog.'
scene_context:
theme: light
spacing: comfortable
layout: isolated_card
placement: center
scale: default
instances: 1
guidance: text
clutter: none
difficulty:
difficulty_bucket: mid
tier: L1
axes_ratings:
precision_requirement: 2
target_acquisition: 3
density_choice_interference: 1
depth_layering: 2
feedback_dynamics: 2
semantic_observability: 2
disambiguation_load: 1
justification: Adds a modal confirmation step and small remove icon target compared to a simple upload/clear action.
success_trigger:
human_readable:
- The "Data attachment" uploader has 0 files remaining in its list.
- The removal confirmation dialog was accepted by clicking "Remove".
canonical_predicate:
predicate_type: equals
target_state:
instance: Data attachment
files: []
count: 0
confirmation:
type: dialog
accepted: true
tolerance: null
require_confirm: true
confirm_control: Remove
require_correct_instance: true
target_instance_label_or_id: Data attachment
terminal_condition: task ends when predicate holds
negative_cases:
- Clicking the trash icon but choosing "Cancel" (file remains).
- '"data_sample.csv" remains present after dialog closes.'
- Removing a different file (if multiple files are present).
- Dialog is opened but no button is clicked.
expected_interaction_path:
- Click the trash/remove icon on the data_sample.csv row
- In the dialog, click "Remove"
- Verify the file list is empty
notes: 'Checker hint: bind to data-testid="uploader-data-attachment" and confirm the dialog result (accepted) plus list
empty.'
- id: file_upload_button-mui-T16
name: Upload a file by matching a reference (MUI)
canonical_type: file_upload_button
implementation_source: mui
implementation_variant: 'composite: Button-as-label + reference card'
implementation_component: 'MUI: Button(label)+hidden input'
task_template: match_reference
secondary_template: null
browsergym_goal: Upload the same file that is shown in the Reference card using the uploader labeled "Upload". The task
will finish automatically when done.
ui_copy: 'Upload: choose the file shown in the Reference card.'
setup_description: 'A centered isolated card titled "Submit supporting file" contains a single MUI Button labeled "Upload"
implemented as a label with a hidden file input. To the right, a Reference card displays the target file''s name and an
icon (the reference is non-interactive). The instruction relies on the reference (visual/mixed guidance) rather than embedding
the filename in the prompt. After selection, a line of text under the button shows "Selected: <filename>".'
scene_context:
theme: light
spacing: comfortable
layout: isolated_card
placement: center
scale: default
instances: 1
guidance: visual
clutter: none
difficulty:
difficulty_bucket: mid
tier: L1
axes_ratings:
precision_requirement: 2
target_acquisition: 2
density_choice_interference: 2
depth_layering: 1
feedback_dynamics: 1
semantic_observability: 2
disambiguation_load: 1
justification: Requires reading the target from the on-page reference card, then performing a standard single-file upload.
success_trigger:
human_readable:
- The uploader labeled "Upload" has exactly one selected file.
- The selected file's name matches the filename displayed in the Reference card.
canonical_predicate:
predicate_type: matches_reference
target_state:
instance: Upload
reference_slot: Reference card
match_field: file_name
count: 1
tolerance: null
require_confirm: false
confirm_control: null
require_correct_instance: true
target_instance_label_or_id: Upload
terminal_condition: task ends when predicate holds
negative_cases:
- Selecting a file that does not match the Reference card.
- Selecting multiple files.
- Only opening the file picker without selecting a file.
expected_interaction_path:
- Read the filename from the Reference card
- Click the "Upload" button
- Select that file
- Verify the selected filename matches
notes: 'Checker hint: reference filename in data-testid="reference-file-name"; compare to input.files[0].name for data-testid="uploader-upload".'
- id: file_upload_button-mui-T17
name: Open the file preview dialog for an uploaded image
canonical_type: file_upload_button
implementation_source: mui
implementation_variant: 'composite: upload + preview dialog'
implementation_component: 'MUI: Button(label)+hidden input + Dialog (preview)'
task_template: open_overlay
secondary_template: null
browsergym_goal: Open the preview dialog for the file "screenshot.webp" in the uploader card and leave it open. The task
will finish automatically when done.
ui_copy: Preview screenshot.webp (open the preview dialog).
setup_description: A centered isolated card titled "Screenshot" contains an upload button (MUI Button-as-label) and a single
pre-selected image file named "screenshot.webp". Next to the filename is a small text button labeled "Preview". Clicking
"Preview" opens a MUI Dialog with the title "Preview" and shows the image along with the filename. The dialog has a "Close"
button but the task requires leaving the preview open. No other dialogs or uploaders exist on the page.
scene_context:
theme: light
spacing: comfortable
layout: isolated_card
placement: center
scale: default
instances: 1
guidance: text
clutter: none
difficulty:
difficulty_bucket: mid
tier: L1
axes_ratings:
precision_requirement: 1
target_acquisition: 2
density_choice_interference: 1
depth_layering: 2
feedback_dynamics: 1
semantic_observability: 2
disambiguation_load: 1
justification: Adds one overlay step (opening a preview dialog) with clear UI affordance and no file selection required.
success_trigger:
human_readable:
- A preview dialog is open (visible) for the uploader card.
- The dialog content corresponds to the file named "screenshot.webp" (filename shown in the dialog).
canonical_predicate:
predicate_type: equals
target_state:
overlay_open: true
overlay_type: dialog
overlay_file_name: screenshot.webp
tolerance: null
require_confirm: false
confirm_control: null
require_correct_instance: false
target_instance_label_or_id: null
terminal_condition: task ends when predicate holds
negative_cases:
- The preview dialog is not open.
- A different overlay is open or the dialog shows a different file.
- The dialog is opened and then closed before the terminal condition is met.
expected_interaction_path:
- Locate the row showing screenshot.webp
- Click the "Preview" button
- Ensure the dialog stays open and shows screenshot.webp
notes: 'Checker hint: check Dialog open state and its filename text; attach data-testid="preview-dialog" with data-file-name="screenshot.webp".'
- id: file_upload_button-mui-T18
name: Upload an exact set of 3 similar-named PDFs (dark, compact, small)
canonical_type: file_upload_button
implementation_source: mui
implementation_variant: 'composite: Button(label)+hidden input (multiple)'
implementation_component: 'MUI: Button (component="label") + visually hidden <input type="file" multiple>'
task_template: file_upload
secondary_template: null
browsergym_goal: 'Using the uploader labeled "Supporting documents", upload exactly these three files: "appendix_a.pdf",
"appendix_b.pdf", and "appendix_c.pdf". The task will finish automatically when done.'
ui_copy: 'Supporting documents: upload appendix_a.pdf, appendix_b.pdf, appendix_c.pdf.'
setup_description: A compact, small-scale upload card is centered on a dark-themed page. The card is titled "Supporting
documents" and contains a single MUI Button-as-label uploader configured with multiple=true and accept="application/pdf".
Below the button is a dense list of selected files where each row shows a filename and a small remove icon. The list starts
empty. Once files are chosen, each should appear as its own row; the component considers the task complete when the list
contains exactly the required three PDFs.
scene_context:
theme: dark
spacing: compact
layout: isolated_card
placement: center
scale: small
instances: 1
guidance: text
clutter: none
difficulty:
difficulty_bucket: hard
tier: L2
axes_ratings:
precision_requirement: 4
target_acquisition: 2
density_choice_interference: 4
depth_layering: 1
feedback_dynamics: 3
semantic_observability: 2
disambiguation_load: 2
justification: Selecting an exact set of multiple files with highly similar names increases precision and interference,
especially in compact/small dark UI.
success_trigger:
human_readable:
- The Supporting documents uploader contains the set of files {appendix_a.pdf, appendix_b.pdf, appendix_c.pdf} and no
others.
- All three files are present in the rendered list (one row per file).
canonical_predicate:
predicate_type: set_membership
target_state:
files:
- appendix_a.pdf
- appendix_b.pdf
- appendix_c.pdf
count: 3
tolerance:
order: any
require_confirm: false
confirm_control: null
require_correct_instance: true
target_instance_label_or_id: Supporting documents
terminal_condition: task ends when predicate holds
negative_cases:
- Only a subset of the three files is uploaded.
- An extra file is uploaded in addition to the three required ones.
- A similarly named but incorrect file is uploaded (e.g., appendix_d.pdf).
- Files appear in UI but underlying input.files does not match exactly.
expected_interaction_path:
- Click the "Supporting documents" upload button
- Select appendix_a.pdf, appendix_b.pdf, and appendix_c.pdf (multi-select)
- Verify exactly these three files appear in the list
notes: 'Checker hint: compare normalized set(input.files[].name) to target set; use data-testid="uploader-supporting-docs".'
- id: file_upload_button-mui-T19
name: Replace a file in a table row uploader (3 instances, table cell)
canonical_type: file_upload_button
implementation_source: mui
implementation_variant: 'composite: table cell uploaders'
implementation_component: 'MUI: Table + Button(label)+hidden input in cells'
task_template: table_operation
secondary_template: file_manage
browsergym_goal: In the "Required documents" table, replace the file in the "Contract" row so that it contains "contract_signed.pdf"
(and not "contract_old.pdf"). The task will finish automatically when done.
ui_copy: 'Required documents table: Contract row should have contract_signed.pdf.'
setup_description: 'The page is a table-cell layout: a MUI Table titled "Required documents" is the main content and occupies
most of the viewport (high clutter due to many rows and columns). Each of three relevant rows (Contract, ID, W-9) contains
an upload control in the rightmost cell implemented as a small MUI Button-as-label with a hidden file input. The Contract
row is pre-filled and currently shows "contract_old.pdf" as the selected file, with a small replace/upload button next
to it. The other rows are either empty or have different placeholder statuses and serve as distractors. The user must
update only the Contract row so that the selected file becomes "contract_signed.pdf".'
scene_context:
theme: light
spacing: comfortable
layout: table_cell
placement: center
scale: default
instances: 3
guidance: text
clutter: high
difficulty:
difficulty_bucket: hard
tier: L2
axes_ratings:
precision_requirement: 3
target_acquisition: 4
density_choice_interference: 3
depth_layering: 1
feedback_dynamics: 2
semantic_observability: 2
disambiguation_load: 4
justification: Dense table layout with multiple similar per-row uploaders increases target acquisition and disambiguation;
replacement requires ensuring the old file is no longer selected.
success_trigger:
human_readable:
- In the "Contract" row uploader, the selected file is exactly "contract_signed.pdf" (count 1).
- '"contract_old.pdf" is not selected in the Contract row.'
- Other rows' uploaders are unchanged.
canonical_predicate:
predicate_type: equals
target_state:
instance: Contract row
files:
- contract_signed.pdf
count: 1
not_present:
- contract_old.pdf
other_instances_unchanged:
ID row: {}
W-9 row: {}
tolerance: null
require_confirm: false
confirm_control: null
require_correct_instance: true
target_instance_label_or_id: Contract row
terminal_condition: task ends when predicate holds
negative_cases:
- Uploading the file to a different row (ID or W-9).
- Both contract_old.pdf and contract_signed.pdf are present/selected.
- The Contract row still shows contract_old.pdf.
- Multiple files are selected in the Contract row.
expected_interaction_path:
- Locate the "Contract" row in the table
- Use its upload/replace control
- Select contract_signed.pdf
- Verify the Contract row shows contract_signed.pdf and not contract_old.pdf
notes: 'Checker hint: each row uploader should have a stable data-testid like uploader-contract, uploader-id, uploader-w9
for instance scoping.'
- id: file_upload_button-mui-T20
name: Remove one specific file from a long scrollable list (bottom-right)
canonical_type: file_upload_button
implementation_source: mui
implementation_variant: 'composite: multi-file list with scroll'
implementation_component: 'MUI: Button(label)+hidden input + scrollable List'
task_template: file_manage
secondary_template: null
browsergym_goal: In the uploader labeled "Logs", remove only the file "log_2025-11-20.txt" from the list and leave all other
files untouched. The task will finish automatically when done.
ui_copy: 'Logs uploader: remove log_2025-11-20.txt only.'
setup_description: 'The upload UI is placed near the bottom-right of the viewport inside a dashboard widget titled "Logs"
(medium clutter: other widgets are visible but not relevant). The uploader supports multiple files and shows an already
populated scrollable MUI List of many text log files (around 12–20 items) with very similar names and dates. Each list
row shows a filename and a small trailing IconButton (trash icon) for removal. The list is constrained to a fixed height,
so scrolling is required to reach some filenames. Initial state includes a row named "log_2025-11-20.txt" somewhere in
the list.'
scene_context:
theme: light
spacing: comfortable
layout: dashboard
placement: bottom_right
scale: default
instances: 1
guidance: text
clutter: medium
difficulty:
difficulty_bucket: hard
tier: L3
axes_ratings:
precision_requirement: 3
target_acquisition: 4
density_choice_interference: 4
depth_layering: 1
feedback_dynamics: 2
semantic_observability: 2
disambiguation_load: 2
justification: Requires finding a specific item among many similar filenames in a scrollable list and clicking a small
remove icon without deleting the wrong file.
success_trigger:
human_readable:
- In the "Logs" uploader, the filename "log_2025-11-20.txt" is no longer present in the file list.
- All other initial files remain present (no additional deletions).
canonical_predicate:
predicate_type: equals
target_state:
instance: Logs
not_present:
- log_2025-11-20.txt
preserve_others: true
tolerance: null
require_confirm: false
confirm_control: null
require_correct_instance: true
target_instance_label_or_id: Logs
terminal_condition: task ends when predicate holds
negative_cases:
- The target file is still present.
- A different log file is removed instead of the target.
- Multiple files are removed (only one should be removed).
- The list becomes empty or changes in any way beyond removing the specified file.
expected_interaction_path:
- Scroll within the Logs file list until log_2025-11-20.txt is visible
- Click the trash/remove icon on that row
- Verify only that row is gone and others remain
notes: 'Checker hint: capture initial file set at task start, then assert final set == initial_set - {target}. data-testid="uploader-logs".'
- id: file_upload_button-mantine-T21
name: Upload a resume with Mantine FileButton
canonical_type: file_upload_button
implementation_source: mantine
implementation_variant: FileButton
implementation_component: 'Mantine: FileButton'
task_template: file_upload
secondary_template: null
browsergym_goal: Upload the file "resume.pdf" using the button labeled "Upload resume". The task will finish automatically
when done.
ui_copy: 'Upload resume: choose resume.pdf.'
setup_description: 'A centered isolated card titled "Resume" contains a Mantine FileButton wrapping a Mantine Button labeled
"Upload resume". The FileButton opens the browser file picker and accepts a single PDF file. Below the button, a text
line reads "Picked file: <name>" once a selection is made (otherwise it is hidden). Initial state: no file picked and
no additional controls are present.'
scene_context:
theme: light
spacing: comfortable
layout: isolated_card
placement: center
scale: default
instances: 1
guidance: text
clutter: none
difficulty:
difficulty_bucket: easy
tier: L0
axes_ratings:
precision_requirement: 2
target_acquisition: 1
density_choice_interference: 1
depth_layering: 1
feedback_dynamics: 1
semantic_observability: 1
disambiguation_load: 1
justification: Single FileButton with straightforward single-file selection and clear textual feedback.
success_trigger:
human_readable:
- The FileButton-controlled input contains exactly one file named "resume.pdf".
- The on-page text shows that the picked file is resume.pdf.
canonical_predicate:
predicate_type: equals
target_state:
files:
- resume.pdf
count: 1
tolerance: null
require_confirm: false
confirm_control: null
require_correct_instance: false
target_instance_label_or_id: null
terminal_condition: task ends when predicate holds
negative_cases:
- A different file is selected.
- No file is selected.
- Multiple files are selected (should be disallowed).
- The UI text updates but the selected file does not match resume.pdf.
expected_interaction_path:
- Click "Upload resume"
- Select resume.pdf
- Verify the picked file text shows resume.pdf
notes: 'Checker hint: bind to data-testid="uploader-resume" and read the stored File object name from component state.'
- id: file_upload_button-mantine-T22
name: Upload a profile photo with Mantine FileInput
canonical_type: file_upload_button
implementation_source: mantine
implementation_variant: FileInput
implementation_component: 'Mantine: FileInput'
task_template: file_upload
secondary_template: null
browsergym_goal: Upload the image file "profile_photo.png" using the input labeled "Profile photo". The task will finish
automatically when done.
ui_copy: 'Profile photo: upload profile_photo.png.'
setup_description: 'A centered isolated card titled "Profile photo" contains a Mantine FileInput with an explicit visible
label "Profile photo" and placeholder text "Select image". The FileInput is configured with accept="image/png,image/jpeg"
and single-file mode (multiple=false). When a file is chosen, the input displays the filename inside the input field.
Initial state: the input is empty (no selected file), and there is no clear button in this task.'
scene_context:
theme: light
spacing: comfortable
layout: isolated_card
placement: center
scale: default
instances: 1
guidance: text
clutter: none
difficulty:
difficulty_bucket: easy
tier: L0
axes_ratings:
precision_requirement: 2
target_acquisition: 2
density_choice_interference: 1
depth_layering: 1
feedback_dynamics: 1
semantic_observability: 1
disambiguation_load: 1
justification: Single labeled FileInput with direct in-field filename display and no additional steps.
success_trigger:
human_readable:
- The FileInput labeled "Profile photo" has exactly one selected file named "profile_photo.png".
- The displayed value shows profile_photo.png.
canonical_predicate:
predicate_type: equals
target_state:
instance: Profile photo
files:
- profile_photo.png
count: 1
tolerance: null
require_confirm: false
confirm_control: null
require_correct_instance: true
target_instance_label_or_id: Profile photo
terminal_condition: task ends when predicate holds
negative_cases:
- A different image file is selected.
- No file is selected.
- A non-image file is selected and rejected.
- The file appears selected but the displayed value does not match.
expected_interaction_path:
- Click the FileInput field or its button area
- Select profile_photo.png
- Verify the filename appears in the input
notes: 'Checker hint: data-testid="uploader-profile-photo" should expose selected file name via state or input.files[0].name.'
- id: file_upload_button-mantine-T23
name: Clear a selected file with FileInput clear button
canonical_type: file_upload_button
implementation_source: mantine
implementation_variant: FileInput (clearable)
implementation_component: 'Mantine: FileInput (clearable)'
task_template: clear_reset
secondary_template: null
browsergym_goal: Clear the selected file in the input labeled "Attachment" so that it becomes empty. The task will finish
automatically when done.
ui_copy: 'Attachment: clear the selected file (currently notes.txt).'
setup_description: 'A centered isolated card titled "Attachment" contains a Mantine FileInput with label "Attachment" and
the prop clearable enabled. The page loads with a file already selected: "notes.txt" is displayed as the current value
inside the input. Because clearable is enabled, a small clear (×) button appears in the right section of the input when
a file is selected. Clicking the clear button resets the FileInput back to its placeholder state with no file selected.'
scene_context:
theme: light
spacing: comfortable
layout: isolated_card
placement: center
scale: default
instances: 1
guidance: text
clutter: none
difficulty:
difficulty_bucket: easy
tier: L0
axes_ratings:
precision_requirement: 2
target_acquisition: 2
density_choice_interference: 1
depth_layering: 1
feedback_dynamics: 1
semantic_observability: 1
disambiguation_load: 1
justification: Single clearable control with an obvious reset affordance and immediate visual state change.
success_trigger:
human_readable:
- The FileInput labeled "Attachment" has no selected file (0 files).
- The input displays its empty/placeholder state (notes.txt is not shown).
canonical_predicate:
predicate_type: equals
target_state:
instance: Attachment
files: []
count: 0
tolerance: null
require_confirm: false
confirm_control: null
require_correct_instance: true
target_instance_label_or_id: Attachment
terminal_condition: task ends when predicate holds
negative_cases:
- notes.txt remains selected.
- A different file remains selected (not cleared).
- The UI appears empty but the underlying file input still contains a file (stale value).
expected_interaction_path:
- Locate the clear (×) button inside the Attachment FileInput
- Click the clear button
- Verify the input is empty
notes: 'Checker hint: for Mantine FileInput, validate value==null (single) for data-testid="uploader-attachment".'
- id: file_upload_button-mantine-T24
name: Upload to the correct FileInput in a form (2 instances)
canonical_type: file_upload_button
implementation_source: mantine
implementation_variant: FileInput (two instances)
implementation_component: 'Mantine: FileInput'
task_template: file_upload
secondary_template: null
browsergym_goal: In the form section, upload the file "id_front.jpg" using the input labeled "ID front" (not "ID back").
The task will finish automatically when done.
ui_copy: 'ID front: upload id_front.jpg.'
setup_description: 'The page is a simple identity form section with low clutter (a few text inputs plus file inputs). There
are two Mantine FileInput components stacked vertically with labels "ID front" and "ID back". Both are single-file inputs
that accept images and display the selected filename in the field. Initial state: both FileInputs are empty (no file selected).
Only the FileInput labeled "ID front" should be filled for this task.'
scene_context:
theme: light
spacing: comfortable
layout: form_section
placement: center
scale: default
instances: 2
guidance: text
clutter: low
difficulty:
difficulty_bucket: mid
tier: L1
axes_ratings:
precision_requirement: 2
target_acquisition: 2
density_choice_interference: 1
depth_layering: 1
feedback_dynamics: 1
semantic_observability: 1
disambiguation_load: 3
justification: Two similar FileInputs require correct instance selection; otherwise the interaction is straightforward
with clear in-field feedback.
success_trigger:
human_readable:
- The FileInput labeled "ID front" has exactly one selected file named "id_front.jpg".
- The FileInput labeled "ID back" remains empty.
canonical_predicate:
predicate_type: equals
target_state:
instance: ID front
files:
- id_front.jpg
count: 1
other_instances_unchanged:
ID back:
count: 0
tolerance: null
require_confirm: false
confirm_control: null
require_correct_instance: true
target_instance_label_or_id: ID front
terminal_condition: task ends when predicate holds
negative_cases:
- Uploading id_front.jpg into the ID back field (wrong instance).
- Selecting a different image file.
- Both fields are filled.
expected_interaction_path:
- Locate the FileInput labeled "ID front"
- Click and select id_front.jpg
- Verify the filename appears in the ID front field only
notes: 'Instrumentation: data-testid="uploader-id-front" and data-testid="uploader-id-back" for instance binding.'
- id: file_upload_button-mantine-T25
name: Match a reference file using Mantine FileButton (top-right)
canonical_type: file_upload_button
implementation_source: mantine
implementation_variant: FileButton + reference card
implementation_component: 'Mantine: FileButton'
task_template: match_reference
secondary_template: null
browsergym_goal: Upload the same file that is shown in the Reference card using the "Upload" button. The task will finish
automatically when done.
ui_copy: 'Upload: choose the file shown in the Reference card.'
setup_description: 'The upload UI is placed near the top-right corner of the viewport. An isolated card titled "Attach file"
contains a Mantine FileButton wrapping a Button labeled "Upload". Next to it, a Reference card displays the target filename
and a small file icon; the reference card is not interactive. After a selection, the page shows a text line under the
button: "Picked file: <name>". The prompt uses the reference card for visual guidance rather than directly specifying
the filename.'
scene_context:
theme: light
spacing: comfortable
layout: isolated_card
placement: top_right
scale: default
instances: 1
guidance: visual
clutter: none
difficulty:
difficulty_bucket: mid
tier: L1
axes_ratings:
precision_requirement: 2
target_acquisition: 1
density_choice_interference: 2
depth_layering: 1
feedback_dynamics: 1
semantic_observability: 2
disambiguation_load: 1
justification: The primary challenge is extracting the target from the reference panel; interaction itself is a simple
FileButton upload.
success_trigger:
human_readable:
- Exactly one file is selected via the FileButton.
- The selected file's name matches the filename shown in the Reference card.
canonical_predicate:
predicate_type: matches_reference
target_state:
reference_slot: Reference card
match_field: file_name
count: 1
tolerance: null
require_confirm: false
confirm_control: null
require_correct_instance: false
target_instance_label_or_id: null
terminal_condition: task ends when predicate holds
negative_cases:
- Selecting a file that does not match the reference filename.
- Selecting multiple files.
- No file selected.
expected_interaction_path:
- Read the filename from the Reference card
- Click the "Upload" button
- Select the referenced file
- Verify the picked file name matches
notes: 'Checker hint: store reference filename in data-testid="reference-file-name" and compare to selected file name stored
in component state.'
- id: file_upload_button-mantine-T26
name: Confirm file removal in a Mantine Modal
canonical_type: file_upload_button
implementation_source: mantine
implementation_variant: FileInput + Modal confirm on remove
implementation_component: 'Mantine: FileInput + Modal'
task_template: confirm_cancel
secondary_template: file_manage
browsergym_goal: Remove the selected file "budget.xlsx" from the input labeled "Attachment" and confirm by clicking "Remove"
in the modal. The task will finish automatically when done.
ui_copy: 'Attachment: remove budget.xlsx (confirm Remove).'
setup_description: 'A centered isolated card titled "Attachment" contains a Mantine FileInput labeled "Attachment" that
is pre-filled with the file "budget.xlsx". To the right of the FileInput is a small outlined button labeled "Remove".
Clicking the Remove button opens a Mantine Modal titled "Remove file?" with two actions: "Cancel" and "Remove". The file
is only cleared from the FileInput after confirming with the modal''s "Remove" button.'
scene_context:
theme: light
spacing: comfortable
layout: isolated_card
placement: center
scale: default
instances: 1
guidance: text
clutter: none
difficulty:
difficulty_bucket: mid
tier: L1
axes_ratings:
precision_requirement: 2
target_acquisition: 2
density_choice_interference: 1
depth_layering: 2
feedback_dynamics: 2
semantic_observability: 2
disambiguation_load: 1
justification: Modal confirmation adds layering and requires correctly choosing the confirm action rather than canceling.
success_trigger:
human_readable:
- The FileInput labeled "Attachment" has no selected file (0 files).
- The confirmation modal was accepted by clicking "Remove".
canonical_predicate:
predicate_type: equals
target_state:
instance: Attachment
files: []
count: 0
confirmation:
type: modal
accepted: true
tolerance: null
require_confirm: true
confirm_control: Remove
require_correct_instance: true
target_instance_label_or_id: Attachment
terminal_condition: task ends when predicate holds
negative_cases:
- Canceling the modal (file remains selected).
- The file remains selected after the modal closes.
- Removing the file without confirming (should not be possible in this setup).
- Closing the modal via outside click/escape without confirming.
expected_interaction_path:
- Click the "Remove" button next to the FileInput
- In the modal, click "Remove"
- Verify the FileInput is empty
notes: 'Checker hint: attach data-testid="remove-modal" with a boolean open state and ensure the file value is null after
confirmation.'
- id: file_upload_button-mantine-T27
name: Drag-and-drop an image into Mantine Dropzone (dark theme)
canonical_type: file_upload_button
implementation_source: mantine
implementation_variant: Dropzone
implementation_component: 'Mantine: Dropzone (@mantine/dropzone)'
task_template: drag_operation
secondary_template: file_upload
browsergym_goal: Drag and drop the image file "screenshot.webp" into the dropzone area. The task will finish automatically
when done.
ui_copy: 'Dropzone: drag screenshot.webp here (or click to select).'
setup_description: 'A centered isolated card on a dark-themed page contains a Mantine Dropzone component with a large bordered
area. The dropzone shows the instructional text "Drag images here or click to select files" and accepts image files. When
a file is dragged over the dropzone, its visual state changes (accept/reject indicators). After a successful drop, the
page renders a small list under the dropzone titled "Accepted files" showing the uploaded filenames. Initial state: no
accepted files are listed.'
scene_context:
theme: dark
spacing: comfortable
layout: isolated_card
placement: center
scale: default
instances: 1
guidance: text
clutter: none
difficulty:
difficulty_bucket: hard
tier: L2
axes_ratings:
precision_requirement: 2
target_acquisition: 3
density_choice_interference: 2
depth_layering: 1
feedback_dynamics: 3
semantic_observability: 2
disambiguation_load: 1
justification: Requires a drag-and-drop interaction to the correct target region; Dropzone state changes and acceptance
feedback add dynamics, especially in dark theme.
success_trigger:
human_readable:
- The Dropzone accepted-files list contains exactly one file named "screenshot.webp".
- The accepted file was provided via a drop interaction (drop event observed).
canonical_predicate:
predicate_type: equals
target_state:
files:
- screenshot.webp
count: 1
interaction_method: drop
accepted: true
tolerance: null
require_confirm: false
confirm_control: null
require_correct_instance: false
target_instance_label_or_id: null
terminal_condition: task ends when predicate holds
negative_cases:
- No file appears in the accepted list.
- A different file is accepted.
- The file is rejected (wrong type/too large) and appears only in a rejected list.
- The file is selected via click-to-select if the task run enforces drop-only.
expected_interaction_path:
- Drag screenshot.webp onto the dropzone area
- Release to drop
- Verify screenshot.webp appears under Accepted files
notes: 'Checker hint: instrument Dropzone onDrop to record interaction_method="drop" and store accepted filenames in a data-testid="dropzone-accepted"
list.'
- id: file_upload_button-mantine-T28
name: Upload an exact 4-file set in a compact settings panel
canonical_type: file_upload_button
implementation_source: mantine
implementation_variant: FileInput (multiple)
implementation_component: 'Mantine: FileInput (multiple)'
task_template: file_upload
secondary_template: null
browsergym_goal: 'Using the input labeled "Supporting files", upload exactly these four files: "budget.xlsx", "data_sample.csv",
"config.json", and "notes.txt". The task will finish automatically when done.'
ui_copy: 'Supporting files: upload budget.xlsx, data_sample.csv, config.json, notes.txt.'
setup_description: 'The page uses a settings-panel layout titled "Project settings" with medium clutter (several toggles
and small helper texts are visible but not required). All content is in compact spacing mode, making controls closer together.
The target component is a Mantine FileInput labeled "Supporting files" configured with multiple=true. Selected files are
displayed as a list of pills below the input (each pill shows the filename and has a small remove icon). Initial state:
no files are selected.'
scene_context:
theme: light
spacing: compact
layout: settings_panel
placement: center
scale: default
instances: 1
guidance: text
clutter: medium
difficulty:
difficulty_bucket: hard
tier: L2
axes_ratings:
precision_requirement: 4
target_acquisition: 3
density_choice_interference: 3
depth_layering: 1
feedback_dynamics: 2
semantic_observability: 2
disambiguation_load: 2
justification: Exact multi-file selection with mixed file types plus compact spacing increases precision and confusability
while still requiring only one component.
success_trigger:
human_readable:
- The FileInput labeled "Supporting files" contains exactly the set {budget.xlsx, data_sample.csv, config.json, notes.txt}
and no other files.
- All four filenames are visible in the selected-files pills/list.
canonical_predicate:
predicate_type: set_membership
target_state:
instance: Supporting files
files:
- budget.xlsx
- data_sample.csv
- config.json
- notes.txt
count: 4
tolerance:
order: any
require_confirm: false
confirm_control: null
require_correct_instance: true
target_instance_label_or_id: Supporting files
terminal_condition: task ends when predicate holds
negative_cases:
- Only a subset of the required files is selected.
- An extra file is selected.
- One of the filenames is misspecified (wrong file).
- Pills show the names but the underlying selected file list does not match exactly.
expected_interaction_path:
- Click the Supporting files FileInput
- Select the four required files (multi-select)
- Verify exactly four pills are present with the correct names
notes: 'Checker hint: normalize to set of file names from component state; validate count==4 and names match target set.'
- id: file_upload_button-mantine-T29
name: Clear the Banner file using a small Reset control (3 instances)
canonical_type: file_upload_button
implementation_source: mantine
implementation_variant: FileButton (resetRef pattern, multiple instances)
implementation_component: 'Mantine: FileButton + external Reset control (resetRef)'
task_template: clear_reset
secondary_template: null
browsergym_goal: Clear the selected file in the uploader labeled "Banner image" by using its "Reset" control. Leave the
"Avatar image" and "Document" uploaders unchanged. The task will finish automatically when done.
ui_copy: 'Banner image: reset/clear the selected file (currently banner.png).'
setup_description: 'The page is a compact form section titled "Assets" containing three side-by-side upload controls (same
canonical type) built with Mantine FileButton. Each uploader has a heading label: "Avatar image", "Banner image", and
"Document". Under each label there is a Button labeled "Upload" (FileButton) and a small red "Reset" action (a compact
button or action icon) that clears the selection via the FileButton resetRef pattern. Initial state: Avatar image already
has "avatar.jpg" selected, Banner image already has "banner.png" selected, and Document already has "resume.pdf" selected;
each selection is shown as a text line "Picked file: <name>" under its uploader. The task is only to clear the Banner
image selection; other two must remain unchanged.'
scene_context:
theme: light
spacing: compact
layout: form_section
placement: center
scale: default
instances: 3
guidance: text
clutter: low
difficulty:
difficulty_bucket: hard
tier: L3
axes_ratings:
precision_requirement: 3
target_acquisition: 4
density_choice_interference: 2
depth_layering: 1
feedback_dynamics: 2
semantic_observability: 2
disambiguation_load: 4
justification: Multiple similar instances plus very small per-instance Reset controls create high disambiguation and target
acquisition difficulty.
success_trigger:
human_readable:
- The uploader labeled "Banner image" has no selected file (0 files).
- The uploaders labeled "Avatar image" and "Document" still show their original selected files.
canonical_predicate:
predicate_type: equals
target_state:
instance: Banner image
files: []
count: 0
other_instances_unchanged:
Avatar image:
files:
- avatar.jpg
count: 1
Document:
files:
- resume.pdf
count: 1
tolerance: null
require_confirm: false
confirm_control: null
require_correct_instance: true
target_instance_label_or_id: Banner image
terminal_condition: task ends when predicate holds
negative_cases:
- Banner image still has banner.png selected.
- Avatar image or Document selection changes (cleared or replaced).
- Clearing all uploaders instead of only Banner image.
- Only hiding the text without clearing the underlying file selection.
expected_interaction_path:
- Locate the column labeled "Banner image"
- Click its small "Reset" control
- Verify Banner image shows no picked file while the other two remain unchanged
notes: 'Checker hint: provide data-testid per instance (uploader-avatar, uploader-banner, uploader-document) and verify
state deltas; instrument resetRef-triggered clear.'
- id: file_upload_button-mantine-T30
name: Open file details overlay from a small icon button
canonical_type: file_upload_button
implementation_source: mantine
implementation_variant: FileInput + details modal
implementation_component: 'Mantine: FileInput + Modal'
task_template: open_overlay
secondary_template: null
browsergym_goal: Open the file details overlay for the selected file "budget.xlsx" and leave it open. The task will finish
automatically when done.
ui_copy: 'Budget file: open details for budget.xlsx.'
setup_description: A centered isolated card titled "Budget file" uses a small-scale Mantine FileInput labeled "Budget file".
The FileInput is pre-filled with the selected file "budget.xlsx" and shows the filename in the field. To the right of
the input is a small info/action icon button labeled (via aria-label) "View details". Clicking the icon opens a Mantine
Modal overlay titled "File details" that displays the filename and basic metadata (type/size). The modal includes a Close
button, but the task requires leaving the overlay open.
scene_context:
theme: light
spacing: comfortable
layout: isolated_card
placement: center
scale: small
instances: 1
guidance: text
clutter: none
difficulty:
difficulty_bucket: hard
tier: L2
axes_ratings:
precision_requirement: 1
target_acquisition: 4
density_choice_interference: 1
depth_layering: 2
feedback_dynamics: 1
semantic_observability: 3
disambiguation_load: 1
justification: A small details icon increases target acquisition difficulty; success depends on opening the correct overlay
and verifying the file identity inside it.
success_trigger:
human_readable:
- A modal overlay titled "File details" is open.
- The overlay shows details for the file named "budget.xlsx".
canonical_predicate:
predicate_type: equals
target_state:
overlay_open: true
overlay_type: modal
overlay_title: File details
overlay_file_name: budget.xlsx
tolerance: null
require_confirm: false
confirm_control: null
require_correct_instance: false
target_instance_label_or_id: null
terminal_condition: task ends when predicate holds
negative_cases:
- The details modal is not open.
- A modal is open but it shows a different file.
- The modal is opened and then closed before success is checked.
- The file selection changes (budget.xlsx is replaced) instead of opening details.
expected_interaction_path:
- Locate the "View details" icon next to the FileInput
- Click it to open the modal
- Ensure the modal remains open and mentions budget.xlsx
notes: 'Checker hint: attach data-testid="file-details-modal" and assert it is open with data-file-name="budget.xlsx".'