メニュー
How these AI agent skills are structured and kept up to date. Use when adding, editing, splitting, or reorganizing skills under .agents/skills, or updating the AGENTS.md router.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
SwiftUI and UIKit UI conventions for Home Assistant iOS. Use when building views, choosing SwiftUI vs UIKit, bridging with embeddedInHostingController(), building reusable components in the HADesignSystem package, or following the one-struct-per-file, inline-body, and #Preview rules.
Home Assistant iOS project layout, build setup, and the "World" dependency-injection pattern. Use when starting work in the repo, deciding where code lives across targets (App, Shared, Watch, CarPlay, Extensions), accessing dependencies through the global `Current`, or setting up dependencies and code signing.
Swift code style, linting, logging, assets, and idioms for Home Assistant iOS. Use when writing or formatting Swift, running SwiftFormat/SwiftLint, referencing SF Symbols or Material Design icons, logging, or using the with() helper.
Asynchronous code and server networking conventions. Use when writing async/await, Tasks, actors, or Combine, deciding whether to touch PromiseKit, annotating view models with @MainActor, or calling the Home Assistant server via HAKit.
Localization via SwiftGen-generated L10n accessors and the en.lproj .strings files. Use when adding or changing user-facing strings, working with the L10n/CoreStrings/FrontendStrings tables, or dealing with Lokalise-managed translations.
The cross-platform MagicItem action model shared by Widgets, Watch, CarPlay, and App Shortcuts. Use when adding or changing item types or action overrides, or touching anything that persists MagicItem via Codable.
| name | ha-ios-skill-maintenance |
| description | How these AI agent skills are structured and kept up to date. Use when adding, editing, splitting, or reorganizing skills under .agents/skills, or updating the AGENTS.md router. |
AI agent instructions for this repo live as modular skills under .agents/skills/, with a compact router in AGENTS.md. This keeps startup context small: an agent loads only the skill relevant to the task instead of one large always-loaded file.
AGENTS.md (repo root) is the router. It carries a short project intro and a table that maps each skill to when it should be loaded..agents/skills/<skill-name>/SKILL.md..claude/skills symlinks to ../.agents/skills so Claude Code discovers the skills. Editor-specific instruction files (CLAUDE.md, .cursorrules, .windsurfrules, .github/copilot-instructions.md) remain symlinks to AGENTS.md.Each file starts with YAML frontmatter:
---
name: ha-ios-<topic>
description: <what the skill covers>. Use when <the concrete triggers that should route an agent here>.
---
name must match the directory name.description is the routing trigger. Write it as "what it covers. Use when …" and name concrete actions/APIs an agent would be doing, so the right skill loads without being told which one.SKILL.md, then add a row to the routing table in AGENTS.md.