apple-dev-skills

Skills for vibe coding iOS & Apple-ecosystem apps with Claude Code — and for driving the agent that builds them.

Languages: English · 繁體中文 · 简体中文 · 日本語

apple-dev-skills is one marketplace for independent developers and small teams building a new, greenfield app on the iOS 26 / macOS 26 floor — from idea to a shipped App Store release. Its first-party skills come in two halves: Apple/Swift skills for what you are building, and harness engineering skills (how you run Claude Code itself — dispatch, review, ship). Every skill is an opinionated default backed by a shipped-it war story and held to a consistency gate. Alongside them sit best-of-breed external skill plugins, aggregated by reference — not written here, linked and credited to their authors, never copied.

Quickstart

Skills trigger automatically from what you ask for — install, then just describe the task.

Inside a Claude Code session, run:

/plugin marketplace add wei18/apple-dev-skills
/plugin install apple-dev-skills@apple-dev-skills          # 26 Apple/Swift skills
/plugin install collaboration-skills@apple-dev-skills      # 12 harness-engineering skills

Each install opens a details pane — pick User scope. The form is plugin@marketplace; the marketplace happens to share its first plugin's name.

If the install summary says Run /reload-plugins to activate., Claude Code runs it for you; if it warns that your next message would re-read the conversation, run /reload-plugins --force.

Install either or both. Externals install the same way, e.g. /plugin install swiftui-expert@apple-dev-skills.

Then just describe the task:

• "I'm starting a new iOS app" → apple-platform-targets answers the deployment target and hands off to the package shape, language mode, and test baseline in order.
• "App Review bounced this on guideline 5.1.2" → app-store-review-rejections maps the guideline to the fix.

To force one, use its slash command: /apple-dev-skills:swift6-concurrency. /skills lists what is installed and which plugin each skill came from.

Installing both plugins loads all 38 first-party descriptions — over 5,000 tokens, roughly 3× the default 1% skill-listing budget (2,000 tokens on a 200k-context model) — before any externals. The skill-listing budget is 1% of the context window (~10,000 tokens on a 1M-context model, where the two first-party plugins alone fit — but externals and your own skills share that same budget); when it overflows, Claude Code drops descriptions starting with the skills you invoke least; in practice that tends to mean freshly installed ones, which have no invocation history yet. Run /doctor to check the listing's cost, then raise skillListingBudgetFraction (e.g. 0.04 on a 200k-context model — these 38 names + descriptions alone need just over 0.03) if it's tight, or disable plugins you don't use via /plugin.

Catalog

• Spec the flow → spec-phase-orchestration
• Bootstrap the project → apple-platform-targets
• Build the UI → swiftui-navigation-architecture
• Test it → swift-testing-baseline
• Ship it → asc-api-automation
• Operate it → apple-three-piece-analytics

Full index in the tables below.

apple-dev-skills (26) — Apple/Swift

Skill	One-liner
swift6-concurrency	Swift 6 language mode + complete concurrency checking; MainActor by default, Sendable only where code crosses into nonisolated/actor context
apple-platform-targets	Default iOS 26 / macOS 26, Xcode 26.x; drop to iOS 18 / macOS 15 (or 17 / 14) only when an older user base requires it
swiftpm-modularization	Single Package, multi-target, thin App, DI composition root, one-to-one tests
swift-testing-baseline	swift-testing + pointfreeco snapshot; protocol fakes; strict/tolerant snapshot gate
xcode-cloud-single-track-ci	Single-track Xcode Cloud; PR / Main / Release / Periodic; pre-merge PR CI
local-archive-export-upload	Local xcodebuild archive → export → altool upload to TestFlight when Xcode Cloud can't run
mise-tool-management	Pin CLI tool versions (swiftlint, xcbeautify…) with mise, so local dev and CI use the exact same ones
oslog-logger-defaults	Default logging setup: Apple's own os.Logger, no third-party library, log values private unless you opt in
apple-three-piece-analytics	App Store Connect (ASC) Analytics + MetricKit + Game Center; no third-party tracking; PrivacyInfo mandatory
telemetry-facade-pattern	One observe(event) call fanned out to OSLog / tracking / Game Center sinks, with MetricKit payloads fed in as events — swap sinks without touching call sites
ai-translated-localization	Default 7 locales; AI translation flow; Localizable.xcstrings; completeness gates
ios-accessibility-engineering	VoiceOver / Dynamic Type / touch-target / Reduce Motion for SwiftUI & UIKit; WCAG 2.2
swift-dependency-injection	Make services swappable for tests — protocol injection + a composition root (environment vs constructor, @TaskLocal, Sendable)
ios-performance-engineering	Instruments / xctrace / hang-hitch budgets / launch / memory / binary size / MetricKit
apple-public-repo-security	Three lines of defence for public iOS/macOS repos + rotate-first leak SOP
build-time-secret-injection	xcconfig + Info.plist $() + Bundle.main for ship-in-binary-but-out-of-diff IDs
storekit2-iap-defaults	StoreKit 2 non-consumable IAP defaults; bridge-protocol test seam, entitlements, restore
monetization-sdk-integration	Add/upgrade/audit a monetization SDK; isolate import to one bridge file
app-store-review-rejections	Diagnose & pre-empt App Review rejection classes for free + ads + IAP + CloudKit + Game Center (GC)
asc-api-automation	ES256 JWT from the .p8 + curl against the ASC REST API — TestFlight, metadata, submission, reports; no fastlane
swiftui-interaction-footguns	Known SwiftUI interaction bugs that slip past pure-code review
swiftui-navigation-architecture	Typed-route navigation for SwiftUI — one @Observable router, NavigationStack, deep links, macOS fallbacks handled
ios-design-mockup	Single-file HTML iOS design mockup from a spec — iPhone frames + tokens
interactive-simulator-ux-audit	Drive a booted Simulator with idb (tap/describe/screenshot) to catch nav/modal/safe-area bugs snapshots can't
host-driven-xcuitest-e2e	Launch-the-app XCUITest E2E via Tuist — dedicated scheme wiring + macOS window-frame click driving
cloudkit-schema-source-of-truth	Committed .ckdb + cktool export/validate/import to Development; Production is a user-owned Console-only gate

collaboration-skills (12) — harness engineering: dispatch, review, ship

Skill	One-liner
spec-phase-orchestration	Pre-implementation doc pipeline; section-by-section approval
subagent-review-cycles	Leader / Developer / Code-Reviewer triad; how many rounds, what counts as a rejection
leader-developer-handoff-contract	6 required elements when dispatching a sub-agent
agent-impl-notes-log†	Running impl-notes during a sub-agent task — decisions, deviations, open questions
subagent-conflict-detection	Check a new sub-agent's targets don't overlap an in-flight worktree
methodology-pattern-extractor†	Extract patterns recurring ≥3 times from meeting logs
session-to-meeting-log†	Consolidate a Claude Code session into a meeting log; summary, not verbatim
pr-diff-verification	Before push/PR, verify git show --stat --summary HEAD matches the commit's claims
backlog-routing-by-topic†	Route stray ideas by topic to the matching spec file's §Backlog
claude-skill-plugin-packaging	Distribute/install Claude Code skills — depth-1 rule, plugin + marketplace, install pinned per project, aggregate by reference
skill-authoring-patterns	Apple/Swift catalog layer over superpowers:writing-skills — router descriptions, bookend sections, two-tier references, evidence-based CR
github-contribution-workflow	gh-CLI contribution loop — PRs, issues, GitHub file ops, secrets, contribution-flow repo settings; conventions

† Assumes the spec-phase-orchestration doc layout (design.md / plan.md / meetings/).

Aggregated external (7) — by reference, credited

Listed here but not authored here: each installs from its author's own repo (you get their latest) and is credited in full. Aggregate, don't appropriate — only MIT-compatible, non-duplicate plugins are listed, and only for genuine gaps. That check happens once, at listing time, not on every upstream commit, so an external's scope and licence can drift afterwards (caveman already has). Externals are broad reference — "here's the API, here's how to build X." First-party skills are narrower: one opinionated default per topic (iOS 26 as the floor, one Package, swift-testing + snapshot, OSLog only, known runtime bugs to avoid). Where a topic overlaps, the two aren't duplicates — they answer at a different level of detail.

Plugin	Author	Covers
apple-skills	Prisma Labs (vabole), MIT	Broad Apple frameworks — SwiftUI, SwiftData, App Intents, WidgetKit, StoreKit, HealthKit, apple-aso, hig …; a SwiftUI performance audit guide that complements ios-performance-engineering's Instruments/MetricKit measurement; also ships simulator-utils and xcuitest (see below for how those line up against this catalog's Simulator/XCUITest skills). A handful of its skills sit in disabled-skills/ — kept in the repo but not loaded by agents
swiftui-expert	Antoine van der Lee (MIT)	SwiftUI patterns, Swift Charts, Liquid Glass, Instruments toolchain
swiftui-pro	Paul Hudson (MIT)	SwiftUI pitfalls, deprecated-API watchlist, iOS 26 / Liquid Glass
caveman	JuliusBrussee (MIT)	Ultra-compressed communication mode — cuts ~65% of output tokens (author's measurement); ships SessionStart/UserPromptSubmit hooks (needs node on PATH, runs every prompt) (general agent behavior)
ponytail	DietrichGebert (MIT)	"Lazy senior dev" mode — forces the simplest, shortest solution (general agent behavior)
i-have-adhd	Ayoub G. (MIT)	Always-on ADHD-friendly output mode — numbered steps, state restated each turn (general agent behavior)
xcode-build-skill	pz (MIT)	xcodebuild/xcrun simctl CLI cheatsheet — schemes → simulators → build → install → launch → screenshot

caveman's license: skills MIT; repo also ships a BSL-1.1 engine. caveman has since grown into a 20-skill suite; 4 of them (caveman-discover, caveman-manage, caveman-evidence-review, caveman-setup) document the author's hosted Caveman Cloud commercial service.

Unlike caveman (token compression) and ponytail (solution simplicity), i-have-adhd shapes the structure of every answer.

xcode-build-skill (xcodebuild/simctl loop) and apple-skills:simulator-utils (simctl one-offs) are CLI references; interactive-simulator-ux-audit drives idb for exploratory audits; host-driven-xcuitest-e2e wires Tuist schemes for CI — apple-skills:xcuitest is the API reference it assumes.

apple-skills moved from the vabole personal account to the Prisma-Labs-Dev organization; this table lists the org's repo.

Other ways to install

(A is the Quickstart above.)

B — pinned for a team

Pin the marketplace to a released tag directly in .claude/settings.json — no submodule needed:

{
  "extraKnownMarketplaces": {
    "apple-dev-skills": {
      "source": { "source": "github", "repo": "wei18/apple-dev-skills", "ref": "v2.1.0" }
    }
  },
  "enabledPlugins": {
    "apple-dev-skills@apple-dev-skills": true,
    "collaboration-skills@apple-dev-skills": true
  }
}

Commit it — collaborators get the marketplace automatically once they trust the project folder, no separate prompt. The two first-party plugins are relative-path entries, so they're installed at that point; externals are GitHub-sourced, so enabledPlugins won't auto-install them for anyone else — each collaborator still runs the claude plugin install … line Claude Code prints, once.

C — npx skills (flat, no plugin)

npx skills add wei18/apple-dev-skills --list
npx skills add wei18/apple-dev-skills --skill swift6-concurrency

Path C does not include the aggregated externals. npx skills reads this repo's marketplace.json / plugin.json, but it only follows locally-declared skill paths. It does not fetch the externals' remote github / git-subdir sources. So the commands above install only the 38 first-party skills — the 7 externals are silently skipped. To flat-install the whole catalog (externals included, pulled from their authors' repos):

scripts/install-flat.sh -g          # user-level; drop -g for project-level
scripts/install-flat.sh --dry-run   # preview the `npx skills add` commands

Contributing

Three ways to help: aggregate an external plugin, add a first-party skill, or report a field note — see CONTRIBUTING.md.

Provenance

First-party skills were distilled and genericized from wei18/Sudoku's .claude/skills/ — a spec-first, AI-Leader/Developer-built portfolio of shipping Apple-platform games — plus original write-ups of public Apple / WCAG / Swift standards. Aggregated externals remain their authors' work, surfaced by reference only. The design specs and plans that produced this repo's two-plugin shape lived under docs/superpowers/ — retired in favour of git history; run git log -- docs/ to find them. MIT — see LICENSE.