SwiftUI Adaptivity Check

Validate existing UI against its supported device/configuration matrix. Owns the adaptivity workflow and completion criteria used independently and by figma-full-cycle stage 6.

Inputs and setup

• Required: project/screens or native captures, with supported configuration boundaries.
• Optional: selected states, device/configuration overrides, design references, output path, permission to fix findings, pipeline context.
• Default report: docs/<RUN>_ADAPTIVITY.md.

Read validation-context.md first: it defines modes, audit, discovery, physical checks, fix scope and reports. <shared-root> is the sibling _figma-full-cycle-shared. Standalone use needs no manifest, UI Plan, Figma URL, Figma credential or mock screenshot.

Load the bundled SwiftUI Adaptive Layout skill by its exact registered name in review mode. Apply LAY-11/12/13 from its canonical layout rules. LAY-11 owns device bounds, Preview/evidence requirements, edge-invariant tolerance and minimum manual coverage. Reuse supplied constraints or derive them from project docs, the brief and relevant views. Unknown support boundaries are a coverage question; one reference device cannot represent the app.

Workflow

1. Build/label the required matrix from deployment target, screen states and resolved Layout, Typography, Strings and Accessibility constraints. Alongside LAY-11 devices include light/dark, supported content sizes, documented locales/directions and long copy, supported orientations, keyboard and safe-area cases. VoiceOver, Reduce Motion and increased contrast require an active Accessibility requirement.
2. Bring screens up directly and capture every required row. Keep geometry-bearing screens' smallest/largest Previews required by LAY-11; adding them needs existing fix authorization, otherwise report missing required Previews as a finding. Supplied captures cover only their documented rows. A minimal available device supplies partial evidence and never closes unavailable required configurations. Use shared Preview evidence for grouped fixtures, destination selection and supported overrides. Resizable Canvas and alternative-display previews can explore supported configurations, but do not replace LAY-11's native captures or prove runtime display transitions. Unsupported required overrides need a different capture route or an open row, not a default-device substitute.
3. Choose evidence per row:

   • Actual design/mock available: use recorded provenance/pin and run the shared helper when sips is available. Only rows with real references enter the batch plan:

     python3 <shared-root>/scripts/screenshot_diff.py batch \
       --plan <artifact-root>/reports/matrix.json --project-root <project-root> \
       --out <artifact-root>/reports/matrix-result.json

     Items are {label, render, mock, ignore_top, ignore_bottom}. Record actual markers/JSON. Follow shared Figma acquisition only when fetching Figma evidence; local design captures need no Figma call.

   • No design/mock, or no sips: mark geometry not-run with its reason. Inspect every row against native bounds/project invariants. Do not fabricate a design from another render, stretch a reference-device capture into a mock, or emit a mechanical clean marker. Direct review is a complete supported validation path.
4. Apply LAY-11 manual coverage to mechanically checked rows. Its cross-aspect clipping limits remain binding; inspect truncation/reflow, overlap and control placement. Verify intended edge relations on smallest/largest targets and the system-owned root shape, including oversized decoration, under shared rules.
5. For pending translations use sanctioned pseudolocales/fixtures or retain pending coverage/an approved waiver under project policy. Do not manufacture translations. Track source-layout success separately from translation readiness.
6. Write the shared report with the full required matrix, including unexecuted rows, measured edge distances, reviewed evidence and fixes/rechecks. Authorized UI changes invalidate affected comparison/adaptivity/physical results; return these rows to the orchestrator in pipeline mode. Return report path, result and coverage gaps.

Completion criteria

• Every required row has reviewed evidence: UI fits, no unintended letterboxing, clipping or overlap, and background reaches system-provided edges.
• LAY-11 edge-invariant, Preview and manual-review requirements are met; unavailable required configurations stay open rather than being replaced by an available device.
• Mechanical rows contain actual markers/JSON; no-reference/no-helper rows name the fallback and have direct review for every row.
• Translation readiness is explicit and required physical checks meet the shared contract.
• A passing result has no required row failed, blocked, not-run or stale. A skipped mechanical pass can coexist with a passed row after its required direct review.