Design Md Enforcer

Design MD Enforcer

Purpose

The design-md-enforcer skill ensures that both human developers and AI agents follow a unified design system specification called DESIGN.md.

It serves two primary functions:

1. Extraction: Analyzing the visible design language of a public URL and converting it into an AI-agent-friendly design system package.
2. Enforcement: Enforcing a system where design reasoning and raw values (hex codes, sizes) are bound together using token roles, preventing drift between style guides and code.

Core Philosophy: Logic Meets Values

1. Tokens are Roles, Not Just Variables:

A token is not just a hex code; it represents a specific purpose or role in the design system.

• Example: primary is not just #0A1F1C; it is the "main ink, text color" role.
• Example: body_main is not just 18px; it is the "default body copy" role.

1. Components Reference Roles:

UI components must never hardcode static hex values. They must reference the token roles. If the role's value changes, all components referencing it are automatically updated.

1. Reasoning is Required (Thematic Storytelling):

Every design decision must be accompanied by its reasoning and a thematic style reference. For example, don't just list colors; describe the overall aesthetic (e.g., "> Midnight Command Center. Imagine a high-tech dashboard glowing softly in a dark room...").

When To Use This Skill

Use this skill when the user asks to:

• Extract a website's design language from a URL.
• Create a DESIGN.md specification from a reference site or for a new project.
• Enforce design rules when an AI agent or developer is creating new UI components.
• Validate color contrasts and design standards (e.g., WCAG).
• Refactor a tailwind or CSS config into a proper DESIGN.md unified spec.

Resource Usage

This skill contains pre-built templates and examples to help you generate accurate outputs. Always use them as your primary reference:

• Read skills/design-md-enforcer/templates/ for the exact structure of DESIGN.md, preview.html, and token files.
• Review skills/design-md-enforcer/examples/ if you are unsure how a high-quality output should look.

Workflow 1: Extracting Design from a URL

When asked to extract a design from a URL, follow these steps:

1. Collect Evidence (Non-Browser LLM Support)

If you do not have native browser capabilities (like a visual subagent), you MUST use alternative methods:

• Fetch the raw HTML and extract inline styles, CSS variable declarations (:root), and class names.
• Ask the user to provide screenshots of the website and use vision models to extract hex codes, fonts, and border radii.
• Ask the user to paste the CSS file content.
• Be explicit about what was observed vs inferred. If you cannot render the site visually, state the limitations.

2. Output Generation Rules (Strict Pathing)

Do not output files directly to the root workspace. Always generate the extracted design system inside a dedicated standard folder: /design-md//

You MUST produce the following minimum files in this folder:

• /design-md//DESIGN.md: The core specification using detailed markdown tables and reasoning.
• /design-md//preview.html: A standalone HTML file demonstrating the extracted components (Hero section, Button variants, Cards, Form controls, Color swatches, etc.) so the user can visually verify the system.

Workflow 2: Enforcing the Specification

1. Structure of DESIGN.md

When generating or updating a DESIGN.md file, you MUST use detailed Markdown tables and incorporate thematic reasoning.

Theme / Style Reference Example:

# AuthKit — Style Reference
> Midnight Command Center. Imagine a high-tech dashboard glowing softly in a dark room, with frosted glass elements reflecting minimal light.

**Theme:** dark

Tokens Table Example: Do not use simple lists for tokens in the body. Use a strict Markdown table format with Name | Value | Role.

## Colors

| Name | Value | Role |
|---|---|---|
| Midnight Abyss | `#05060f` | Page backgrounds, elevated card backgrounds, deep shadows. |
| Ghost White | `#ffffff` | Primary text for headings and high-contrast elements, icon fills. |
| Storm Gray | `#2f343` | Subtle shadows, secondary background for interactive states. |
| Comet | `#d8ecf8` | Primary body text, prominent links, and headings; high readability. |
| Celestial Light | `#b6d9fc` | Focus states for interactive elements, subtle highlights. |

Components Block: Use YAML (or standard markdown code blocks) to define components inheriting these tokens.

components:
  button_primary:
    background: "{tokens.colors.Comet}"
    text: "{tokens.colors.Midnight Abyss}"

2. Validation & Linting Requirements

When you or another agent create or modify design tokens and components, you MUST validate the decisions:

• Contrast Check: Verify that text colors over background colors pass WCAG accessibility standards (minimum 4.5:1 for normal text).
• Token Integrity: Ensure that components only reference existing tokens defined in the DESIGN.md tables.
• Role Consistency: Check that the assigned token role makes logical sense for the component.

If a validation fails, you must catch the error, correct the color or token reference, and explain the correction.

3. Component Hierarchy & Overrides

When defining components, promote the use of inheritance (extends). Instead of redefining every property for a hover state, extend the base component and only override the properties that change.

Agent Behavior

• Never hardcode hex colors or px sizes directly in HTML/CSS/Tailwind classes if a token exists for it.
• If a token does not exist, propose adding it to DESIGN.md along with its reasoning and role before using it in a component.
• Always read DESIGN.md before making any UI changes to understand the established roles and reasoning.
• Act as the strict gatekeeper: reject any UI code generation that violates the DESIGN.md specification.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: fatih-developer
• Source: fatih-developer/fth-skills
• License: MIT

Install and usage instructions live in the source repository linked above.