fix(desktop): realign real-window smoke diagnostic with Base UI Dialog modal

[Astro-Han]

What

The search modal migrated to the Base UI Dialog primitives, which silently dropped the selectors the real-window smoke diagnostic relied on:

• the backdrop stopped carrying .maka-search-modal-backdrop
• the focused input lost the maka-search-modal-input class to InputGroup's own utility classes

As a result the programmatic-search-modal-open and programmatic-focus-target smoke checks failed on a clean tree (reproduced identically on main). This is stale harness, not a product regression — the modal renders and traps focus correctly; only the diagnostic's selectors had drifted.

Changes

• packages/ui/src/ui.tsx — give the shared DialogBackdrop a style-free maka-dialog-backdrop hook so the diagnostic can select the backdrop reliably (Base UI otherwise renders only drifting utility classes).
• apps/desktop/src/main/main.ts — query .maka-dialog-backdrop, and detect the focus trap structurally via closest('.maka-search-modal') instead of a brittle class on the focused element.
• scripts/desktop-real-window-smoke.mjs — focus check accepts any interactive control (input or button) trapped inside the modal.
• apps/desktop/src/renderer/styles.css — drop the dead .maka-search-modal-backdrop rule (no element carries it after the Dialog migration).
• tests — add smoke-diagnostic-contract.test.ts binding the diagnostic to the stable hook so it can't drift again, and realign the existing real-window-smoke-contract assertion to the structural focus check.

Verification

• npm --workspace @maka/desktop run smoke:programmatic-window → 6/6 checks PASS (previously 2 FAILed).
• npm --workspace @maka/desktop test → 1528/1529 pass.

Status — rebased onto post-#79 main

The previously-noted app-icon failure was a pre-existing main breakage owned by #79. That PR has merged (squash 989632f8) and this PR is now rebased onto the updated main, so the icon assertion passes. The full desktop suite is green here except for one flaky, unrelated test (rive-workflow-tool → "kills and reaps a Rive child that ignores SIGTERM on abort") — a timing-sensitive subprocess-reap check that passes on isolated re-run (3/3). This PR stays scoped to the smoke-harness drift and touches neither the icon nor the Rive code.

https://claude.ai/code/session_01VWXzHoyncpGYuwbaDEDizD

[Astro-Han]

What

The search modal migrated to the Base UI Dialog primitives, which silently dropped the selectors the real-window smoke diagnostic relied on:

• the backdrop stopped carrying .maka-search-modal-backdrop
• the focused input lost the maka-search-modal-input class to InputGroup's own utility classes

As a result the programmatic-search-modal-open and programmatic-focus-target smoke checks failed on a clean tree (reproduced identically on main). This is stale harness, not a product regression — the modal renders and traps focus correctly; only the diagnostic's selectors had drifted.

Changes

• packages/ui/src/ui.tsx — give the shared DialogBackdrop a style-free maka-dialog-backdrop hook so the diagnostic can select the backdrop reliably (Base UI otherwise renders only drifting utility classes).
• apps/desktop/src/main/main.ts — query .maka-dialog-backdrop, and detect the focus trap structurally via closest('.maka-search-modal') instead of a brittle class on the focused element.
• scripts/desktop-real-window-smoke.mjs — focus check accepts any interactive control (input or button) trapped inside the modal.
• apps/desktop/src/renderer/styles.css — drop the dead .maka-search-modal-backdrop rule (no element carries it after the Dialog migration).
• tests — add smoke-diagnostic-contract.test.ts binding the diagnostic to the stable hook so it can't drift again, and realign the existing real-window-smoke-contract assertion to the structural focus check.

Verification

• npm --workspace @maka/desktop run smoke:programmatic-window → 6/6 checks PASS (previously 2 FAILed).
• npm --workspace @maka/desktop test → 1528/1529 pass.

Status — rebased onto post-#79 main

The previously-noted app-icon failure was a pre-existing main breakage owned by #79. That PR has merged (squash 989632f8) and this PR is now rebased onto the updated main, so the icon assertion passes. The full desktop suite is green here except for one flaky, unrelated test (rive-workflow-tool → "kills and reaps a Rive child that ignores SIGTERM on abort") — a timing-sensitive subprocess-reap check that passes on isolated re-run (3/3). This PR stays scoped to the smoke-harness drift and touches neither the icon nor the Rive code.

https://claude.ai/code/session_01VWXzHoyncpGYuwbaDEDizD

[Astro-Han]

What

The search modal migrated to the Base UI Dialog primitives, which silently dropped the selectors the real-window smoke diagnostic relied on:

• the backdrop stopped carrying .maka-search-modal-backdrop
• the focused input lost the maka-search-modal-input class to InputGroup's own utility classes

As a result the programmatic-search-modal-open and programmatic-focus-target smoke checks failed on a clean tree (reproduced identically on main). This is stale harness, not a product regression — the modal renders and traps focus correctly; only the diagnostic's selectors had drifted.

Changes

• packages/ui/src/ui.tsx — give the shared DialogBackdrop a style-free maka-dialog-backdrop hook so the diagnostic can select the backdrop reliably (Base UI otherwise renders only drifting utility classes).
• apps/desktop/src/main/main.ts — query .maka-dialog-backdrop, and detect the focus trap structurally via closest('.maka-search-modal') instead of a brittle class on the focused element.
• scripts/desktop-real-window-smoke.mjs — focus check accepts any interactive control (input or button) trapped inside the modal.
• apps/desktop/src/renderer/styles.css — drop the dead .maka-search-modal-backdrop rule (no element carries it after the Dialog migration).
• tests — add smoke-diagnostic-contract.test.ts binding the diagnostic to the stable hook so it can't drift again, and realign the existing real-window-smoke-contract assertion to the structural focus check.

Verification

• npm --workspace @maka/desktop run smoke:programmatic-window → 6/6 checks PASS (previously 2 FAILed).
• npm --workspace @maka/desktop test → 1528/1529 pass.

Status — rebased onto post-#79 main

The previously-noted app-icon failure was a pre-existing main breakage owned by #79. That PR has merged (squash 989632f8) and this PR is now rebased onto the updated main, so the icon assertion passes. The full desktop suite is green here except for one flaky, unrelated test (rive-workflow-tool → "kills and reaps a Rive child that ignores SIGTERM on abort") — a timing-sensitive subprocess-reap check that passes on isolated re-run (3/3). This PR stays scoped to the smoke-harness drift and touches neither the icon nor the Rive code.

https://claude.ai/code/session_01VWXzHoyncpGYuwbaDEDizD

[Astro-Han]

What

The search modal migrated to the Base UI Dialog primitives, which silently dropped the selectors the real-window smoke diagnostic relied on:

• the backdrop stopped carrying .maka-search-modal-backdrop
• the focused input lost the maka-search-modal-input class to InputGroup's own utility classes

As a result the programmatic-search-modal-open and programmatic-focus-target smoke checks failed on a clean tree (reproduced identically on main). This is stale harness, not a product regression — the modal renders and traps focus correctly; only the diagnostic's selectors had drifted.

Changes

• packages/ui/src/ui.tsx — give the shared DialogBackdrop a style-free maka-dialog-backdrop hook so the diagnostic can select the backdrop reliably (Base UI otherwise renders only drifting utility classes).
• apps/desktop/src/main/main.ts — query .maka-dialog-backdrop, and detect the focus trap structurally via closest('.maka-search-modal') instead of a brittle class on the focused element.
• scripts/desktop-real-window-smoke.mjs — focus check accepts any interactive control (input or button) trapped inside the modal.
• apps/desktop/src/renderer/styles.css — drop the dead .maka-search-modal-backdrop rule (no element carries it after the Dialog migration).
• tests — add smoke-diagnostic-contract.test.ts binding the diagnostic to the stable hook so it can't drift again, and realign the existing real-window-smoke-contract assertion to the structural focus check.

Verification

• npm --workspace @maka/desktop run smoke:programmatic-window → 6/6 checks PASS (previously 2 FAILed).
• npm --workspace @maka/desktop test → 1528/1529 pass.

Status — rebased onto post-#79 main

The previously-noted app-icon failure was a pre-existing main breakage owned by #79. That PR has merged (squash 989632f8) and this PR is now rebased onto the updated main, so the icon assertion passes. The full desktop suite is green here except for one flaky, unrelated test (rive-workflow-tool → "kills and reaps a Rive child that ignores SIGTERM on abort") — a timing-sensitive subprocess-reap check that passes on isolated re-run (3/3). This PR stays scoped to the smoke-harness drift and touches neither the icon nor the Rive code.

https://claude.ai/code/session_01VWXzHoyncpGYuwbaDEDizD

[Astro-Han]

What

The search modal migrated to the Base UI Dialog primitives, which silently dropped the selectors the real-window smoke diagnostic relied on:

• the backdrop stopped carrying .maka-search-modal-backdrop
• the focused input lost the maka-search-modal-input class to InputGroup's own utility classes

As a result the programmatic-search-modal-open and programmatic-focus-target smoke checks failed on a clean tree (reproduced identically on main). This is stale harness, not a product regression — the modal renders and traps focus correctly; only the diagnostic's selectors had drifted.

Changes

• packages/ui/src/ui.tsx — give the shared DialogBackdrop a style-free maka-dialog-backdrop hook so the diagnostic can select the backdrop reliably (Base UI otherwise renders only drifting utility classes).
• apps/desktop/src/main/main.ts — query .maka-dialog-backdrop, and detect the focus trap structurally via closest('.maka-search-modal') instead of a brittle class on the focused element.
• scripts/desktop-real-window-smoke.mjs — focus check accepts any interactive control (input or button) trapped inside the modal.
• apps/desktop/src/renderer/styles.css — drop the dead .maka-search-modal-backdrop rule (no element carries it after the Dialog migration).
• tests — add smoke-diagnostic-contract.test.ts binding the diagnostic to the stable hook so it can't drift again, and realign the existing real-window-smoke-contract assertion to the structural focus check.

Verification

• npm --workspace @maka/desktop run smoke:programmatic-window → 6/6 checks PASS (previously 2 FAILed).
• npm --workspace @maka/desktop test → 1528/1529 pass.

Status — rebased onto post-#79 main

The previously-noted app-icon failure was a pre-existing main breakage owned by #79. That PR has merged (squash 989632f8) and this PR is now rebased onto the updated main, so the icon assertion passes. The full desktop suite is green here except for one flaky, unrelated test (rive-workflow-tool → "kills and reaps a Rive child that ignores SIGTERM on abort") — a timing-sensitive subprocess-reap check that passes on isolated re-run (3/3). This PR stays scoped to the smoke-harness drift and touches neither the icon nor the Rive code.

https://claude.ai/code/session_01VWXzHoyncpGYuwbaDEDizD

[Astro-Han]

What

The search modal migrated to the Base UI Dialog primitives, which silently dropped the selectors the real-window smoke diagnostic relied on:

• the backdrop stopped carrying .maka-search-modal-backdrop
• the focused input lost the maka-search-modal-input class to InputGroup's own utility classes

As a result the programmatic-search-modal-open and programmatic-focus-target smoke checks failed on a clean tree (reproduced identically on main). This is stale harness, not a product regression — the modal renders and traps focus correctly; only the diagnostic's selectors had drifted.

Changes

• packages/ui/src/ui.tsx — give the shared DialogBackdrop a style-free maka-dialog-backdrop hook so the diagnostic can select the backdrop reliably (Base UI otherwise renders only drifting utility classes).
• apps/desktop/src/main/main.ts — query .maka-dialog-backdrop, and detect the focus trap structurally via closest('.maka-search-modal') instead of a brittle class on the focused element.
• scripts/desktop-real-window-smoke.mjs — focus check accepts any interactive control (input or button) trapped inside the modal.
• apps/desktop/src/renderer/styles.css — drop the dead .maka-search-modal-backdrop rule (no element carries it after the Dialog migration).
• tests — add smoke-diagnostic-contract.test.ts binding the diagnostic to the stable hook so it can't drift again, and realign the existing real-window-smoke-contract assertion to the structural focus check.

Verification

• npm --workspace @maka/desktop run smoke:programmatic-window → 6/6 checks PASS (previously 2 FAILed).
• npm --workspace @maka/desktop test → 1528/1529 pass.

Status — rebased onto post-#79 main

The previously-noted app-icon failure was a pre-existing main breakage owned by #79. That PR has merged (squash 989632f8) and this PR is now rebased onto the updated main, so the icon assertion passes. The full desktop suite is green here except for one flaky, unrelated test (rive-workflow-tool → "kills and reaps a Rive child that ignores SIGTERM on abort") — a timing-sensitive subprocess-reap check that passes on isolated re-run (3/3). This PR stays scoped to the smoke-harness drift and touches neither the icon nor the Rive code.

https://claude.ai/code/session_01VWXzHoyncpGYuwbaDEDizD

[Astro-Han]

What

The search modal migrated to the Base UI Dialog primitives, which silently dropped the selectors the real-window smoke diagnostic relied on:

• the backdrop stopped carrying .maka-search-modal-backdrop
• the focused input lost the maka-search-modal-input class to InputGroup's own utility classes

As a result the programmatic-search-modal-open and programmatic-focus-target smoke checks failed on a clean tree (reproduced identically on main). This is stale harness, not a product regression — the modal renders and traps focus correctly; only the diagnostic's selectors had drifted.

Changes

• packages/ui/src/ui.tsx — give the shared DialogBackdrop a style-free maka-dialog-backdrop hook so the diagnostic can select the backdrop reliably (Base UI otherwise renders only drifting utility classes).
• apps/desktop/src/main/main.ts — query .maka-dialog-backdrop, and detect the focus trap structurally via closest('.maka-search-modal') instead of a brittle class on the focused element.
• scripts/desktop-real-window-smoke.mjs — focus check accepts any interactive control (input or button) trapped inside the modal.
• apps/desktop/src/renderer/styles.css — drop the dead .maka-search-modal-backdrop rule (no element carries it after the Dialog migration).
• tests — add smoke-diagnostic-contract.test.ts binding the diagnostic to the stable hook so it can't drift again, and realign the existing real-window-smoke-contract assertion to the structural focus check.

Verification

• npm --workspace @maka/desktop run smoke:programmatic-window → 6/6 checks PASS (previously 2 FAILed).
• npm --workspace @maka/desktop test → 1528/1529 pass.

Status — rebased onto post-#79 main

The previously-noted app-icon failure was a pre-existing main breakage owned by #79. That PR has merged (squash 989632f8) and this PR is now rebased onto the updated main, so the icon assertion passes. The full desktop suite is green here except for one flaky, unrelated test (rive-workflow-tool → "kills and reaps a Rive child that ignores SIGTERM on abort") — a timing-sensitive subprocess-reap check that passes on isolated re-run (3/3). This PR stays scoped to the smoke-harness drift and touches neither the icon nor the Rive code.

https://claude.ai/code/session_01VWXzHoyncpGYuwbaDEDizD

[Astro-Han]

What

The search modal migrated to the Base UI Dialog primitives, which silently dropped the selectors the real-window smoke diagnostic relied on:

• the backdrop stopped carrying .maka-search-modal-backdrop
• the focused input lost the maka-search-modal-input class to InputGroup's own utility classes

As a result the programmatic-search-modal-open and programmatic-focus-target smoke checks failed on a clean tree (reproduced identically on main). This is stale harness, not a product regression — the modal renders and traps focus correctly; only the diagnostic's selectors had drifted.

Changes

• packages/ui/src/ui.tsx — give the shared DialogBackdrop a style-free maka-dialog-backdrop hook so the diagnostic can select the backdrop reliably (Base UI otherwise renders only drifting utility classes).
• apps/desktop/src/main/main.ts — query .maka-dialog-backdrop, and detect the focus trap structurally via closest('.maka-search-modal') instead of a brittle class on the focused element.
• scripts/desktop-real-window-smoke.mjs — focus check accepts any interactive control (input or button) trapped inside the modal.
• apps/desktop/src/renderer/styles.css — drop the dead .maka-search-modal-backdrop rule (no element carries it after the Dialog migration).
• tests — add smoke-diagnostic-contract.test.ts binding the diagnostic to the stable hook so it can't drift again, and realign the existing real-window-smoke-contract assertion to the structural focus check.

Verification

• npm --workspace @maka/desktop run smoke:programmatic-window → 6/6 checks PASS (previously 2 FAILed).
• npm --workspace @maka/desktop test → 1528/1529 pass.

Status — rebased onto post-#79 main

The previously-noted app-icon failure was a pre-existing main breakage owned by #79. That PR has merged (squash 989632f8) and this PR is now rebased onto the updated main, so the icon assertion passes. The full desktop suite is green here except for one flaky, unrelated test (rive-workflow-tool → "kills and reaps a Rive child that ignores SIGTERM on abort") — a timing-sensitive subprocess-reap check that passes on isolated re-run (3/3). This PR stays scoped to the smoke-harness drift and touches neither the icon nor the Rive code.

https://claude.ai/code/session_01VWXzHoyncpGYuwbaDEDizD