Add SwiftUI Expert eval docs

README.md

@@ -192,8 +192,22 @@ swiftui-expert-skill/
     trace-recording.md
     view-structure.md - View extraction and composition patterns
 ```
+
 <!-- END REFERENCE STRUCTURE -->
 
+## Evals and production telemetry
+
+The `evals/swiftui-expert/` directory contains a small human-review eval set for
+Liquid Glass review, list performance, and Instruments trace analysis. The cases
+are harness-neutral so SwiftUI Expert behavior can be checked before release in
+Claude Code, Codex, Cursor, or another agent workspace.
+
+If you publish this skill through Telvine, keep runtime telemetry metadata-only:
+`skill.invocation.start`, `skill.invocation.end`, and `skill.invocation.error`
+for skill behavior, plus `plugin.component.invoked` and
+`plugin.component.error` for non-skill components. Do not emit prompts, project
+files, trace contents, screenshots, app data, tool arguments, or model outputs.
+
 ## Maintenance
 
 The repository includes a maintenance skill for keeping API guidance current:

evals/swiftui-expert/cases.jsonl

@@ -0,0 +1,3 @@
+{"id":"liquid-glass-review","input":"Review a SwiftUI screen adopting Liquid Glass effects and identify fallback, legibility, and accessibility risks.","expected_outcome":"Uses the SwiftUI Expert references, checks iOS 26+ assumptions, legibility, contrast, accessibility, and fallback behavior."}
+{"id":"performance-list","input":"Analyze a SwiftUI feed with nested ForEach views, inline filtering, AsyncImage thumbnails, and frequent state updates.","expected_outcome":"Flags identity, filtering, image caching/downsampling, observable granularity, and measurement steps before broad rewrites."}
+{"id":"trace-analysis","input":"Analyze a trace summary showing animation hitches and main-thread samples during scrolling.","expected_outcome":"Uses trace-analysis guidance, distinguishes blocked vs CPU-bound main thread, correlates evidence, and proposes targeted fixes."}

evals/swiftui-expert/rubric.md

@@ -0,0 +1,27 @@
+# SwiftUI Expert eval rubric
+
+Score each case from 1-5.
+
+## Reference routing
+
+- 5: Uses the right SwiftUI Expert reference for layout, state, Liquid Glass, performance, accessibility, or trace analysis.
+- 3: Uses relevant guidance but misses a more specific reference.
+- 1: Ignores the skill structure.
+
+## Technical accuracy
+
+- 5: Gives current SwiftUI and platform guidance with clear availability assumptions.
+- 3: Mostly correct with small API or platform gaps.
+- 1: Suggests invalid or deprecated approaches.
+
+## Diagnostic quality
+
+- 5: Connects symptoms to evidence, measurement, and targeted fixes.
+- 3: Gives plausible fixes with weak evidence.
+- 1: Jumps to broad rewrites without diagnosis.
+
+## Privacy and telemetry
+
+- 5: Avoids emitting prompts, project files, trace contents, app data, tool arguments, or model outputs beyond reviewed snippets.
+- 3: Includes unnecessary operational detail without sensitive data.
+- 1: Exposes private app code, traces, or user data.