Review ios ui

Review iOS UI

Dispatch

This skill dispatches 3 specialist agents in parallel:

1. apple-ui-craft:apple-ui-reviewer
   - Scope: all files in the review scope
   - Reviews: HIG, Liquid Glass, typography, color, SF Symbols, navigation, layout, micro-interactions

2. apple-ui-craft:animation-haptics-engineer
   - Scope: same files
   - Reviews: spring parameters, animation purpose, haptic placement, Reduce Motion

3. apple-ui-craft:accessibility-engineer
   - Scope: same files
   - Reviews: VoiceOver, Dynamic Type, contrast, touch targets, motor accessibility

After all 3 complete, merge findings:

• Validate the format first: every finding uses the field names and dimension strings in references/review/01-finding-format.md. Dedup and the verdict table both key on <Dimension>, so a paraphrased dimension becomes an orphan with no verdict row. Re-dispatch a non-conforming specialist rather than repairing its output by hand
• Deduplicate (e.g., touch targets flagged by both accessibility and UI reviewer)
• Order by severity (CRITICAL -> HIGH -> MEDIUM -> LOW -> NIT)
• Group by screen/file
• If two specialists' recommendations conflict, present both with the trade-off named -- never silently pick one
• Present unified report with per-dimension verdicts

Carry NOT ASSESSED through unchanged. apple-ui-reviewer covers four dimensions that need a sequence, a configuration change, or a measurement (task flow, information architecture, error recovery, adaptive layout). When the evidence mode could not show them it returns NOT ASSESSED, and that value belongs in the report verbatim. Promoting it to a clean verdict during the merge turns an honest gap into a false assurance on exactly the class the evidence could not cover.

Scope determination

Arg	Meaning
empty / diff	Uncommitted + staged changes (default)
staged	Only staged files
<file> or <directory>	Specific target
all	Entire project (excluding generated code, Pods, build artifacts)

Output

Unified report with:

• The evidence mode and the coverage it produced (screens reached, states forced, configurations exercised) as the first line -- it bounds every verdict below it
• Per-dimension verdicts: HIG (APPLE-NATIVE/CLOSE/NEEDS WORK/GENERIC), Density and economy (EARNED/ACCEPTABLE/WASTEFUL/STRETCHED-PHONE), Usability and flow (COMPLETABLE/WORKABLE/OBSTRUCTED/BROKEN/NOT ASSESSED), Adaptive layout (ROBUST/ADEQUATE/FRAGILE/BROKEN/NOT ASSESSED), Animation (FLUID/ADEQUATE/STIFF/BROKEN), Haptics (INTENTIONAL/SPARSE/ABSENT/NOISY), Accessibility (INCLUSIVE/ADEQUATE/GAPS/EXCLUDING) -- animation and haptics are two independent verdicts; report both, never collapse them into one label
• Findings organized by file/screen
• Concrete SwiftUI rewrites for every finding
• Top 3 priority actions

All findings are advisory. The user chooses what to apply.

Execution mode

Every agent this skill dispatches is pinned to Opus 5 (model: "opus") at dispatch -- the coding/review floor (owner directive 2026-07-24); the session conductor stays orchestrator-only. When the session model is already the strongest tier and the review scope is small, the orchestrator may run a specialist's review inline in the main context (foreground) instead of dispatching a separate agent, without weakening the read-only guarantee the reviewer agents carry.

Review ledger (write it, without asking)

At the end of every run, write .claude/apple-ui-craft/last-review.json in the REVIEWED repo, and read any existing one first to produce a delta report (NEW / RESOLVED / STILL OPEN / REGRESSED / IMPROVED). Schema, matching rules, and the two fields that stop a narrow run from erasing a wide one (dimensions and notAssessed): references/review/04-run-artifacts.md#review-ledger-schema-v1.

A missing or unreadable ledger is an empty prior run, never an error. A ledger from a different scope is not a prior run for this scope.

Ultracode conductor mode

When the harness announces ultracode, this skill runs conductor-executor per references/_scaffolding/conductor-dispatch-protocol.md -- read that file before the first executor dispatch; it owns the dispatch mechanics, the fan-out doctrine (executor teams scale to natural breadth; the session-model agent caps do not apply to them), the executor prompt contract, and the validation gate. Without ultracode, run the standard 3-specialist dispatch above unchanged.

Split of labor

Conductor (session model -- never delegated)	Conductor-selected executors (Sonnet 5 @ xhigh or Opus 5)
Per-dimension verdicts, severity grading, finding dedup across dimensions, final report synthesis	Per-screen evidence sweeps (one executor per screen group): HIG deviations, contrast pairs, touch-target measurements, Dynamic Type breakpoints -- raw evidence tables for the conductor and the 3 specialists to grade

Executor scoping (on top of the protocol's prompt contract)

• Each executor owns one screen group (non-overlapping) and gets the evidence-table format inline.
• Reference set: absolute paths of the review dimension's reference files + references/_scaffolding/version-floor-registry.md.
• Inline the severity scale (CRITICAL/HIGH/MEDIUM/LOW/NIT) and the 11-row a11y/perf gate from agents/apple-ui-reviewer.md (sourced from references/accessibility/05-motion-accessibility.md, references/patterns/01-gotchas-anti-patterns.md, references/performance/01-swiftui-rendering.md).
• The apple-ui-reviewer / animation-haptics-engineer / accessibility-engineer specialists are pinned to Opus 5 at dispatch -- judgment reviewers, never grunt executors.