justinwetch/HIGAgentSkills

Apple Human Interface Guidelines for Agents

What it is

apple‑hig is a skill (a plug‑in) for conversational AI agents that lets the agent answer questions about Apple’s Human Interface Guidelines (HIG). The repository ships a distilled, source‑reviewed set of 157 HIG references (cut from ~262 k words to ~90 k words) and a special /duo slash‑command that guides an iOS developer through adapting an existing app for the new folding iPhone Duo (iOS 27). The skill does not contain any machine‑learning models; it simply provides curated text and a deterministic workflow that the host agent can invoke.

Core features

Feature What it does
HIG reference corpus 157 topic files (references/hig/*.md) containing the exact wording, values and conditionals from Apple’s official guidelines, rewritten to be much shorter while preserving fidelity.
/duo command A multi‑step workflow (commands/duo.md) that: 1️⃣ assesses an app’s screens in code, 2️⃣ matches them to Apple’s Duo guidance, 3️⃣ presents options with SwiftUI previews, 4️⃣ generates the required SwiftUI/UIKit changes, and 5️⃣ verifies them in the iPhone Duo simulator.
Enforce mode (optional) When asked, the skill can run a “review‑and‑repair” loop with two independent reviewers scoring the UI (90 / 100 pass required). It repeats up to three cycles, fixing issues automatically if the host provides the necessary sub‑agents and artifact access.
Routing helper A small Python script (scripts/hig_route.py) that maps a user request to the appropriate reference file(s) based on literal matches, keeping the token load low.
Packaging A deterministic ZIP builder (scripts/package_runtime_zip.py) creates a runtime bundle that includes only the needed files (SKILL.md, routing-index.md, references/, commands/).

How to install

  1. Copy‑paste prompt – paste the provided text block into your AI‑agent’s chat. The agent will clone the repo, place the apple-hig/ folder in its skill directory, and register the /duo command.
  2. Manual install – download the ZIP, extract the whole apple-hig/ folder into the agent’s skill location, and ensure SKILL.md and the commands/ folder are present.
  3. Slash‑command registration – some agents auto‑detect commands/duo.md; others require you to copy that file into the agent’s custom slash‑command directory and restart the host.
  4. Verify – ask the agent to locate the skill and report its version (2026‑09‑27) and whether /duo is available.

Typical usage

Use apple-hig to review the navigation and accessibility of this iOS app.

The agent will load the relevant foundation files, the platform‑specific HIG sections, and any one‑hop related guidance, then answer with direct quotations from the source clause followed by a concise recommendation.

Duo workflow example

/duo the library and reading flow

The command walks the developer through:

  1. Assess – scans the app’s SwiftUI/Storyboard files.
  2. Match – compares each screen to the Duo checklist.
  3. Decide – shows SwiftUI preview options and asks for confirmation.
  4. Build – writes the necessary code changes.
  5. Verify – runs the app in the iPhone Duo simulator (if available). A duo‑notes.md file is kept in the project so a later /duo continue can pick up where it left off.

Enforce mode example

Use apple-hig in enforce mode to review and fix this interface.

The skill spawns two reviewer sub‑agents, scores the UI, applies fixes, and repeats until the pass threshold is met or three cycles are exhausted.

Technical requirements

  • Python 3.10+ – needed for the routing helper and enforce‑mode scoring.
  • No special hardware; the skill works on any host that can read plain‑text markdown files.
  • The agent must support custom slash‑commands or be able to load files from a skill directory.

License & attribution

  • Code and the distilled reference files are released under the MIT License.
  • The underlying Apple HIG material remains Apple’s property; the repository includes a NOTICE file clarifying that the project is not affiliated with Apple.

This summary is based solely on the repository’s README; no additional features have been inferred.

Related

  • Project
  • Project
  • Project
  • Project
  • Project