Install in Claude Code
Copygit clone --depth 1 https://github.com/Orkas-AI/Orkas /tmp/swiftui-dev && cp -r /tmp/swiftui-dev/resources/builtin/marketplace/skills/b1f384166705 ~/.claude/skills/swiftui-devThen start a new Claude Code session; the skill loads automatically.
Definition
SKILL.md
# SwiftUI Dev Use this skill for SwiftUI development, architecture, structure, performance, and Apple native app profiling. It combines: - SwiftUI view refactor and Observation/MV guidance. - SwiftUI code-first performance audit. - Native macOS/iOS Time Profiler CLI workflow with `xctrace`, `atos`, and included scripts. Do not use this skill for web performance, server profiling, generic PR review, product planning, or non-Apple UI frameworks. For ordinary implementation work without SwiftUI architecture/performance/native profiling concerns, use the development skill. ## Route The Work | User intent | Read | |---|---| | Refactor SwiftUI View structure, split large `body`, review Observation boundaries | [SwiftUI refactor](references/swiftui-refactor.md) | | Decide whether MV, framework-native state, or a ViewModel is justified | [MV patterns](references/mv-patterns.md) | | SwiftUI page is slow, scrolling janks, or body updates too often | [SwiftUI performance](references/swiftui-performance.md) | | Build a dependency/update mental model before deeper profiling | [WWDC23 performance model](references/demystify-swiftui-performance-wwdc23.md) | | Inspect current SwiftUI Instrument lanes or Cause & Effect evidence | [SwiftUI Instruments workflow](references/optimizing-swiftui-performance-instruments.md) and [SwiftUI timeline guide](references/understanding-improving-swiftui-performance.md) | | Diagnose a main-thread hang or run-loop stall | [App hangs guide](references/understanding-hangs-in-your-app.md) | | Record/analyze macOS or iOS native Time Profiler traces from CLI, symbolicate and rank hotspots through the standard Orkas Skill Runner | [Native trace workflow](references/native-trace.md) | If the user provides only symptoms, start with code-first SwiftUI review. Ask for trace/screenshots only when code review is inconclusive or the user explicitly wants trace analysis. ## Required Inputs For SwiftUI code review: - Target view or feature code. - Data flow: `@State`, `@Binding`, `@Environment`, `@Observable`, `@Query`, services, models. - Symptoms and reproduction steps. - Device/OS/build configuration when performance is involved. For native profiling: - App name and binary path, or existing `.trace` bundle. - Attach PID or launch binary path. - Slow interaction to exercise during capture. - Capture duration, defaulting to 90 seconds. - Matching symbols / binary and permission to launch or attach. Proceed with explicit assumptions for code-only reviews. Stop before profiling if required tools or permissions are missing. ## Core Workflow 1. **Classify the task**: SwiftUI structure/refactor, SwiftUI performance, or native trace profiling. 2. **Start code-first for SwiftUI**: inspect structure, data flow, Observation ownership, identity, layout, and body work before requesting heavier tooling. 3. **Use Instruments evidence when available**: treat traces/screenshots/CSV as evidence, not as the whole answer. 4. **Make targeted recommendations**: order issues by likely impact and confidence. 5. **Verify with the same scenario**: compare before/after CPU, frame drops, memory peak, sample counts, or the specific user-visible symptom when data exists. ## Boundaries - Do not invent metrics, trace findings, sample counts, call stacks, device settings, or benchmark results. - Do not run `xctrace` unless the user has provided a local target, trace, or explicit permission and the environment is macOS with Xcode tools. - Do not interpret stale symbols as real hotspots. Binary, trace, and load address must match. - Do not introduce ViewModels by default. Prefer SwiftUI-native state/data flow unless existing code or user requirements justify a view model. - Do not change business logic during refactor unless the user explicitly asks. - If source files must be edited, follow the existing project style and verify with relevant build/tests/profiling. ## Output Format Use this concise structure unless a reference template is more specific: ```markdown # SwiftUI Dev Review: [Target] ## Inputs And Assumptions ## Findings | Priority | Finding | Evidence | Recommendation | Confidence | |---|---|---|---|---| ## Suggested Changes ## Verification Plan ## Risks / Limits ## Handoff To Next Stage ``` For native trace analysis, use the report structure in the [native trace workflow](references/native-trace.md). ## Quality Checklist Before finalizing, verify: - The task is Apple-native or SwiftUI-specific. - SwiftUI structure recommendations preserve clear data ownership and framework-native state flow. - Findings cite code, trace evidence, screenshots, or clearly labeled assumptions. - Recommendations are targeted, not generic "optimize everything" advice. - Refactors preserve behavior unless requested otherwise. - Profiling commands, trace path, binary path, load address, and confidence are reported when CLI profiling is used. - The user gets a concrete verification plan.