Context Summary

Active Context

Currently working on: iOS skills suite — 4 new skills for autonomous iOS development loop
Blocking issues: none
Next up: Parallel implementation of F001-F004

All skills follow the established pattern: agent definition + SKILL.md + references/ + scripts/ + assets/templates/
Shell scripts must pass shellcheck; plists/entitlements must pass plutil -lint
Each skill gets registered in .claude-plugin/marketplace.json
iOS skills must NOT be copies of macOS skills — iOS has fundamentally different testing surfaces, signing model, and interaction patterns

ios-build-verify is P0 because it's the linchpin — without build+simulator feedback, no autonomous loop (2026-02-01)
ios-tdd-expert is P0 because without tests the agent can't verify correctness (2026-02-01)
ios-cicd-distributor is P1 — needed for shipping but the implement→test loop works without it (2026-02-01)
ios-app-architect is P2 — helpful for complex apps but ios-senior-engineer covers basics (2026-02-01)
F001-F004 are fully parallelizable — no shared modules except marketplace.json (2026-02-01)

Skill structure: .claude/agents/NAME.md + NAME/SKILL.md + NAME/references/.md + NAME/scripts/.sh + NAME/assets/templates/*
Agent definitions: model opus, tools Read/Edit/Bash/Grep/Glob/WebSearch/WebFetch
Reference macos-tdd-expert and macos-cicd-distributor for structural templates
Scripts must be parameterized via env vars, idempotent, with clear usage docs

iOS signing uses provisioning profiles (macOS does not) — don't conflate the two models
XCUITest for iOS is touch-based (tap, swipe, scroll) vs macOS (click, type, menu)
performAccessibilityAudit() is iOS 17+ only
ditto not zip for preserving code signatures
set-key-partition-list is required for headless CI signing
Snapshot tests are sensitive to iOS version + device size — must pin in CI