The Problem

Design teams and solo developers who use AI coding tools (Claude Code, Cursor, Codex) have no reliable way to inject consistent design knowledge into AI assistants. Default AI output tends toward generic layouts and arbitrary visual choices. Each tool has its own rules-file format, so sharing design guidance across tools means duplicating and reformatting content.

What This Does

ai-design-skills is a collection of self-contained, markdown-based "skills" that teach AI coding tools how to design. Each skill lives in its own folder under skills/ and contains a single SKILL.md file. The included skills/landing-page-design/SKILL.md covers landing page construction: intake questions, page structure, conversion copy, and a visual system (typography, spacing, radius, motion).

The project is a fork of elayadesign/ai-design-skills (1,081 stars upstream), so the design methodology is battle-tested even though this copy is new. The README documents installation for Claude Code, Cursor, Codex, Windsurf, Cline, and generic system-prompt use.

How It Is Wired

Execution starts with the user asking their AI tool for a design task. The AI reads the installed SKILL.md as a rules file and follows its instructions. There is no application code, no runtime, and no call graph — the entire mechanism is the markdown file being loaded by the AI tool's rules engine.

  • README.md — installation instructions and the skills table; the only navigation aid.
  • skills/landing-page-design/SKILL.md — the actual design logic; the only file that does work. It instructs the AI to ask intake questions, then produce page structure, copy, and a visual system.
  • .gitignore and LICENSE — standard hygiene; MIT license.

The blast radius is the SKILL.md file itself. A change to its design-values section (fonts, colors, spacing) propagates to every AI session that loads it. The wiring beyond that — how each AI tool parses and applies the rules — is external and not mapped in this repo. There is no code to trace, no database, no network calls.

How To Use It

Setup — clone the repo and copy the skill folder into your AI tool's rules directory. The README documents these commands verbatim for Claude Code:

mkdir -p .claude/skills
git clone https://github.com/moses-y/ai-design-skills.git /tmp/ai-design-skills
cp -r /tmp/ai-design-skills/skills/landing-page-design .claude/skills/

For Cursor, the README shows a curl command that fetches the raw SKILL.md into .cursor/rules/. Codex and Windsurf users copy the file into AGENTS.md or .windsurfrules respectively.

Configuration — none required. The design values live in a section near the bottom of SKILL.md; fork the repo and edit that section to customize.

Running it — ask your AI assistant, in plain language, for a landing page design. The AI applies the skill automatically once the rules file is installed.

Real-World Use

A marketing team using Claude Code wants consistent landing pages across campaigns. They install the skill once, then prompt: "Build a landing page for our new analytics product, targeting CTOs." The AI follows the skill's intake questions, produces the page structure and copy, and applies the visual system defined in the skill's design-values section. The team forks the repo, changes the color palette and fonts in that section, and every future page matches their brand.

Code Health & Issues

Static analysis findings (heuristic, verify before acting):

  • Med — No test files detected — untested code paths — repository-wide. For a repo of markdown rules, this is expected; there is no executable code to test. Verification would mean manually confirming the skill produces correct output in each AI tool.
  • Med — No CI/CD pipeline detected — no automated build/test gate — .github/ or CI config. Again, low impact for markdown-only content.

The repo has a license (MIT), no committed secrets, and no lockfile (no dependencies). The main gap is documentation: only one skill exists so far, and the README's "More on the way" promise is unfulfilled.

The Bottom Line

A clean, minimal approach to a real problem: packaging design expertise as portable markdown rules. The single-file-per-skill structure is simple to fork and customize, and the MIT license removes friction. It is a content repo, not a code repo — useful for anyone who wants consistent AI-generated design output, but there is nothing to build, test, or deploy.