# Comparison results: confirm-delete [Comparison overview, scores, protocol and limits](why-no-skill.md) All nine confirm-delete builds from the 2026-10-15 cohort follow unchanged: three arms, three runs each. These checks grade the resulting apps, not operational safety, complete accessibility or physical-device VoiceOver behavior. Read the overview's protocol notes and limits before comparing scores. ## Every run and check ### confirm-delete, shadcn/ui, run 1: 100/100 | Screen | Arm | Score | Typecheck | Lint | Axe | Behavior | | --- | --- | ---: | ---: | ---: | ---: | ---: | | confirm-delete | shadcn | 100 | 20/20 | 10/10 | 20/20 | 50/50 | Every check passed. Problems the app had before the agent ran, which cost nothing: 0 type errors, 0 lint errors. | Behavior check | Acceptance items | Viewport | Result | | --- | --- | --- | --- | | Each cookbook shows its recipe count, and "Delete " opens a dialog titled "Delete ?" with Cancel and Delete cookbook | 1, 2 | desktop | pass | | Focus starts on Cancel | 3 | desktop | pass | | Escape and Cancel close the dialog without deleting, and focus returns to the button that opened it | 3 | desktop | pass | | Delete cookbook deletes once, is disabled or marked busy meanwhile, closes, removes the cookbook and announces it | 4 | desktop | pass | | When the server refuses, its message shows in the dialog, which stays open, and the cookbook stays | 5 | desktop | pass | | At 390px wide the page does not scroll sideways | 6 | phone | pass | | Axe | Result | | --- | --- | | 390px light | pass | | 390px dark | pass | | 1280px light | pass | | 1280px dark | pass | Screenshots, as evidence only: 8 in scripts/agent-usability/runs/2026-10-15/shadcn/confirm-delete/1/score/screenshots. ### confirm-delete, shadcn/ui, run 2: 100/100 | Screen | Arm | Score | Typecheck | Lint | Axe | Behavior | | --- | --- | ---: | ---: | ---: | ---: | ---: | | confirm-delete | shadcn | 100 | 20/20 | 10/10 | 20/20 | 50/50 | Every check passed. Problems the app had before the agent ran, which cost nothing: 0 type errors, 0 lint errors. | Behavior check | Acceptance items | Viewport | Result | | --- | --- | --- | --- | | Each cookbook shows its recipe count, and "Delete " opens a dialog titled "Delete ?" with Cancel and Delete cookbook | 1, 2 | desktop | pass | | Focus starts on Cancel | 3 | desktop | pass | | Escape and Cancel close the dialog without deleting, and focus returns to the button that opened it | 3 | desktop | pass | | Delete cookbook deletes once, is disabled or marked busy meanwhile, closes, removes the cookbook and announces it | 4 | desktop | pass | | When the server refuses, its message shows in the dialog, which stays open, and the cookbook stays | 5 | desktop | pass | | At 390px wide the page does not scroll sideways | 6 | phone | pass | | Axe | Result | | --- | --- | | 390px light | pass | | 390px dark | pass | | 1280px light | pass | | 1280px dark | pass | Screenshots, as evidence only: 8 in scripts/agent-usability/runs/2026-10-15/shadcn/confirm-delete/2/score/screenshots. ### confirm-delete, shadcn/ui, run 3: 100/100 | Screen | Arm | Score | Typecheck | Lint | Axe | Behavior | | --- | --- | ---: | ---: | ---: | ---: | ---: | | confirm-delete | shadcn | 100 | 20/20 | 10/10 | 20/20 | 50/50 | Every check passed. Problems the app had before the agent ran, which cost nothing: 0 type errors, 0 lint errors. | Behavior check | Acceptance items | Viewport | Result | | --- | --- | --- | --- | | Each cookbook shows its recipe count, and "Delete " opens a dialog titled "Delete ?" with Cancel and Delete cookbook | 1, 2 | desktop | pass | | Focus starts on Cancel | 3 | desktop | pass | | Escape and Cancel close the dialog without deleting, and focus returns to the button that opened it | 3 | desktop | pass | | Delete cookbook deletes once, is disabled or marked busy meanwhile, closes, removes the cookbook and announces it | 4 | desktop | pass | | When the server refuses, its message shows in the dialog, which stays open, and the cookbook stays | 5 | desktop | pass | | At 390px wide the page does not scroll sideways | 6 | phone | pass | | Axe | Result | | --- | --- | | 390px light | pass | | 390px dark | pass | | 1280px light | pass | | 1280px dark | pass | Screenshots, as evidence only: 8 in scripts/agent-usability/runs/2026-10-15/shadcn/confirm-delete/3/score/screenshots. ### confirm-delete, shadcn/ui with its skill, run 1: 100/100 | Screen | Arm | Score | Typecheck | Lint | Axe | Behavior | | --- | --- | ---: | ---: | ---: | ---: | ---: | | confirm-delete | shadcn-skill | 100 | 20/20 | 10/10 | 20/20 | 50/50 | Every check passed. Problems the app had before the agent ran, which cost nothing: 0 type errors, 0 lint errors. | Behavior check | Acceptance items | Viewport | Result | | --- | --- | --- | --- | | Each cookbook shows its recipe count, and "Delete " opens a dialog titled "Delete ?" with Cancel and Delete cookbook | 1, 2 | desktop | pass | | Focus starts on Cancel | 3 | desktop | pass | | Escape and Cancel close the dialog without deleting, and focus returns to the button that opened it | 3 | desktop | pass | | Delete cookbook deletes once, is disabled or marked busy meanwhile, closes, removes the cookbook and announces it | 4 | desktop | pass | | When the server refuses, its message shows in the dialog, which stays open, and the cookbook stays | 5 | desktop | pass | | At 390px wide the page does not scroll sideways | 6 | phone | pass | | Axe | Result | | --- | --- | | 390px light | pass | | 390px dark | pass | | 1280px light | pass | | 1280px dark | pass | Screenshots, as evidence only: 8 in scripts/agent-usability/runs/2026-10-15/shadcn-skill/confirm-delete/1/score/screenshots. ### confirm-delete, shadcn/ui with its skill, run 2: 100/100 | Screen | Arm | Score | Typecheck | Lint | Axe | Behavior | | --- | --- | ---: | ---: | ---: | ---: | ---: | | confirm-delete | shadcn-skill | 100 | 20/20 | 10/10 | 20/20 | 50/50 | Every check passed. Problems the app had before the agent ran, which cost nothing: 0 type errors, 0 lint errors. | Behavior check | Acceptance items | Viewport | Result | | --- | --- | --- | --- | | Each cookbook shows its recipe count, and "Delete " opens a dialog titled "Delete ?" with Cancel and Delete cookbook | 1, 2 | desktop | pass | | Focus starts on Cancel | 3 | desktop | pass | | Escape and Cancel close the dialog without deleting, and focus returns to the button that opened it | 3 | desktop | pass | | Delete cookbook deletes once, is disabled or marked busy meanwhile, closes, removes the cookbook and announces it | 4 | desktop | pass | | When the server refuses, its message shows in the dialog, which stays open, and the cookbook stays | 5 | desktop | pass | | At 390px wide the page does not scroll sideways | 6 | phone | pass | | Axe | Result | | --- | --- | | 390px light | pass | | 390px dark | pass | | 1280px light | pass | | 1280px dark | pass | Screenshots, as evidence only: 8 in scripts/agent-usability/runs/2026-10-15/shadcn-skill/confirm-delete/2/score/screenshots. ### confirm-delete, shadcn/ui with its skill, run 3: 100/100 | Screen | Arm | Score | Typecheck | Lint | Axe | Behavior | | --- | --- | ---: | ---: | ---: | ---: | ---: | | confirm-delete | shadcn-skill | 100 | 20/20 | 10/10 | 20/20 | 50/50 | Every check passed. Problems the app had before the agent ran, which cost nothing: 0 type errors, 0 lint errors. | Behavior check | Acceptance items | Viewport | Result | | --- | --- | --- | --- | | Each cookbook shows its recipe count, and "Delete " opens a dialog titled "Delete ?" with Cancel and Delete cookbook | 1, 2 | desktop | pass | | Focus starts on Cancel | 3 | desktop | pass | | Escape and Cancel close the dialog without deleting, and focus returns to the button that opened it | 3 | desktop | pass | | Delete cookbook deletes once, is disabled or marked busy meanwhile, closes, removes the cookbook and announces it | 4 | desktop | pass | | When the server refuses, its message shows in the dialog, which stays open, and the cookbook stays | 5 | desktop | pass | | At 390px wide the page does not scroll sideways | 6 | phone | pass | | Axe | Result | | --- | --- | | 390px light | pass | | 390px dark | pass | | 1280px light | pass | | 1280px dark | pass | Screenshots, as evidence only: 8 in scripts/agent-usability/runs/2026-10-15/shadcn-skill/confirm-delete/3/score/screenshots. ### confirm-delete, Ouro HX, run 1: 100/100 | Screen | Arm | Score | Typecheck | Lint | Axe | Behavior | | --- | --- | ---: | ---: | ---: | ---: | ---: | | confirm-delete | ouro | 100 | 20/20 | 10/10 | 20/20 | 50/50 | Every check passed. Problems the app had before the agent ran, which cost nothing: 0 type errors, 0 lint errors. | Behavior check | Acceptance items | Viewport | Result | | --- | --- | --- | --- | | Each cookbook shows its recipe count, and "Delete " opens a dialog titled "Delete ?" with Cancel and Delete cookbook | 1, 2 | desktop | pass | | Focus starts on Cancel | 3 | desktop | pass | | Escape and Cancel close the dialog without deleting, and focus returns to the button that opened it | 3 | desktop | pass | | Delete cookbook deletes once, is disabled or marked busy meanwhile, closes, removes the cookbook and announces it | 4 | desktop | pass | | When the server refuses, its message shows in the dialog, which stays open, and the cookbook stays | 5 | desktop | pass | | At 390px wide the page does not scroll sideways | 6 | phone | pass | | Axe | Result | | --- | --- | | 390px light | pass | | 390px dark | pass | | 1280px light | pass | | 1280px dark | pass | Screenshots, as evidence only: 8 in scripts/agent-usability/runs/2026-10-15/ouro/confirm-delete/1/score/screenshots. ### confirm-delete, Ouro HX, run 2: 100/100 | Screen | Arm | Score | Typecheck | Lint | Axe | Behavior | | --- | --- | ---: | ---: | ---: | ---: | ---: | | confirm-delete | ouro | 100 | 20/20 | 10/10 | 20/20 | 50/50 | Every check passed. Problems the app had before the agent ran, which cost nothing: 0 type errors, 0 lint errors. | Behavior check | Acceptance items | Viewport | Result | | --- | --- | --- | --- | | Each cookbook shows its recipe count, and "Delete " opens a dialog titled "Delete ?" with Cancel and Delete cookbook | 1, 2 | desktop | pass | | Focus starts on Cancel | 3 | desktop | pass | | Escape and Cancel close the dialog without deleting, and focus returns to the button that opened it | 3 | desktop | pass | | Delete cookbook deletes once, is disabled or marked busy meanwhile, closes, removes the cookbook and announces it | 4 | desktop | pass | | When the server refuses, its message shows in the dialog, which stays open, and the cookbook stays | 5 | desktop | pass | | At 390px wide the page does not scroll sideways | 6 | phone | pass | | Axe | Result | | --- | --- | | 390px light | pass | | 390px dark | pass | | 1280px light | pass | | 1280px dark | pass | Screenshots, as evidence only: 8 in scripts/agent-usability/runs/2026-10-15/ouro/confirm-delete/2/score/screenshots. ### confirm-delete, Ouro HX, run 3: 100/100 | Screen | Arm | Score | Typecheck | Lint | Axe | Behavior | | --- | --- | ---: | ---: | ---: | ---: | ---: | | confirm-delete | ouro | 100 | 20/20 | 10/10 | 20/20 | 50/50 | Every check passed. Problems the app had before the agent ran, which cost nothing: 0 type errors, 0 lint errors. | Behavior check | Acceptance items | Viewport | Result | | --- | --- | --- | --- | | Each cookbook shows its recipe count, and "Delete " opens a dialog titled "Delete ?" with Cancel and Delete cookbook | 1, 2 | desktop | pass | | Focus starts on Cancel | 3 | desktop | pass | | Escape and Cancel close the dialog without deleting, and focus returns to the button that opened it | 3 | desktop | pass | | Delete cookbook deletes once, is disabled or marked busy meanwhile, closes, removes the cookbook and announces it | 4 | desktop | pass | | When the server refuses, its message shows in the dialog, which stays open, and the cookbook stays | 5 | desktop | pass | | At 390px wide the page does not scroll sideways | 6 | phone | pass | | Axe | Result | | --- | --- | | 390px light | pass | | 390px dark | pass | | 1280px light | pass | | 1280px dark | pass | Screenshots, as evidence only: 8 in scripts/agent-usability/runs/2026-10-15/ouro/confirm-delete/3/score/screenshots.