What it does This repo is a structured knowledge pack — not code, but guidance — in the open Agent Skills format. It feeds your AI assistant concise, correct defaults for SwiftUI: which property wrapper to use, how to keep list identity stable, when @Observable beats @StateObject, and what iOS 26 Liquid Glass actually requires. Think of it as a senior dev in a box, minus the coffee breaks.

The interesting bit The standout piece is a bundled Python toolchain that wraps Apple’s xctrace. The agent can record an Instruments trace, then parse five lanes — Time Profiler, Hangs, Animation Hitches, SwiftUI updates, and cause graphs — into correlated JSON and markdown. A single derived metric, main_running_coverage_pct, splits hangs into “blocked on I/O” versus “CPU-bound,” which steers the fix in opposite directions. That’s actual diagnostic logic, not just prompt engineering.

Key highlights

• Covers the full SwiftUI surface: state management, view composition, NavigationStack, Swift Charts, macOS scenes, accessibility, image downsampling, and deprecated-to-modern API migration (iOS 15 through iOS 26+)
• Reference files load on demand so the agent’s context window isn’t bloated with irrelevant topics
• Installable via skills.sh, Claude Code plugin, Cursor plugin, Codex/OpenAI manifest, or manual symlink — six different toolchains supported
• Trace analysis is pure stdlib Python 3; only external dependency is Xcode’s own xctrace
• Includes a maintenance sub-skill for scanning Apple docs and refreshing the deprecated API reference after new OS releases

Caveats

• The maintenance skill requires the Sosumi MCP to auto-scan Apple documentation; without it, updates are manual
• Trace recording only works where xctrace is available (macOS with Xcode installed)
• The repo is guidance and scripts, not a library you import — value depends entirely on your AI tool’s skill support

Verdict Worth a look if you’re shipping SwiftUI and your AI assistant keeps suggesting @ObservedObject where @Observable belongs. Skip it if you don’t use an agent that supports the Skill format, or if you already have a team wiki your LLM is fine-tuned on.