A free toolkit feeds Apple's design rules to AI coding assistants

What was posted

A developer shared a project on Show HN (the Hacker News section where people show off their own work), hosted on GitHub. As github.com describes it, the project gives AI coding agents direct access to Apple's Human Interface Guidelines, or HIG. These are Apple's official rules for how apps should look and behave on its devices.

The project's pitch is that most AI coding assistants rely on what they absorbed during training, which can go stale. The README lists the kinds of mistakes that follows: guessing the size of touch targets, inventing navigation bars Apple has dropped, and ignoring how people use eye gaze on Apple's headset. The project's answer is to fetch the guidelines from Apple's servers when asked.

Three ways to use it

The same engine is packaged three ways, so you can pick whichever fits your setup.

• An agent skill. This is a folder of instructions and reference files that a coding agent can load. It includes Apple's six core design principles, accessibility contrast rules, and pre-written reference pages for each of the guidelines' six main areas.
• A command-line tool. From a terminal you can search the guidelines, print a full topic as readable text, sync a local copy of the reference files, and install the skill into your agent's folders.
• An MCP server. MCP, the Model Context Protocol, is a standard way for AI apps to call outside tools. This server offers four: list sections, search, get a topic, and sync. The README names Gemini CLI, Claude Desktop, Cursor, Windsurf and Zed as compatible hosts, plus any other MCP-compliant app.

How it gets Apple's text

Apple publishes its guidelines through DocC, its documentation format, and exposes the pages as JSON data that anyone can read without logging in. The tool requests those pages directly. It then walks through the page's underlying structure, which the README calls an abstract syntax tree, a nested map of headings, lists, tables, code samples and callout boxes.

Next it converts that structure into plain Markdown, a simple text format that AI models read easily. Links between Apple pages are turned into readable references. The project also swaps typographic dashes for ordinary hyphens, which it says keeps the output consistent from run to run.

The skill's own fetcher script uses only Python's built-in libraries. That means an agent running in a locked-down sandbox can use it without installing anything extra.

What the skill tells an agent

Besides fetching pages, the skill bundles a short checklist of rules. On contrast, it asks for at least a 4.5:1 ratio for normal text and 3:1 for large text, and aims for 7:1 where possible. It says never to signal status by color alone, and to pair color with symbols or text labels.

It also tells agents to use Apple's built-in semantic colors, which adapt to light and dark modes, and to use system text styles rather than fixed pixel sizes. For translucent surfaces such as sheets and sidebars, it points to Apple's system materials, including the Liquid Glass look.

To keep the bundled copy from drifting, a scheduled GitHub workflow runs every Monday at 09:00 UTC. It runs the tests first, then re-syncs and commits any changes Apple has made to its documentation.

How to try it, and what to keep in mind

Setup needs Python 3.12 or newer and uv, a Python package manager. With uv installed, you can run commands straight from the GitHub repository without downloading it first. For the MCP server, you add a short block of settings to your AI app's configuration file, and the README gives examples for Gemini CLI, Claude Desktop and Cursor.

This is an early, community-shared project. The README includes unit tests, but it offers no evidence on how much the tool improves the code an AI writes. Treat it as an experiment, and read the code before connecting it to anything important.