Skip to content

avelikiy/great_cto / skills/anydesign

anydesign

Verified

Analyze images, websites, and Figma files to extract their design and generate a `design.md` with token system, component inventory, and reconstruction notes. Use this skill whenever the user wants to understand, document, replicate, or audit the design of something visual: a screenshot, a URL, a Figma link, a Pinterest reference, a mockup, a competitor's site, a component, a dashboard, a landing page. Also when they ask 'extract the design system from X', 'document the style of Y', 'analyze this visually', 'convert this image into tokens', 'help me replicate this design', 'what palette does this site use', 'how is this built'. Also for single elements: 'copy this navbar', 'recreate this illustration', 'give me a prompt to regenerate this graphic' — element mode outputs a focused element.md, with token-grounded image-model prompts when the element is visual art. If the user brings any visual source and wants to understand it at a design level — this skill should activate.

references/token-extraction.md

View on GitHub
# Token Extraction — How to infer tokens with rigor

This reference defines how to go from "I see colors and sizes" to a documented token system.
It's the difference between describing and systematizing.

---

## Philosophy

**Token = named, reusable design decision.**

A color is not a token; `primary-500 = #3B82F6` with a semantic role is. Your job is to
**infer the system** the designer used (consciously or not), not list loose values.

Three principles:

1. **Look for repetition.** If a value appears once, it's not a token. If it appears three or
   more times, it probably is.
2. **Infer semantic roles, not just numeric scales.** "Dark blue color" is worse than
   "primary". "14px size" is worse than "body-sm".
3. **Mark confidence honestly.** If unsure about the role, say so.

---

## Color tokens

### Process

1. **Identify unique colors** observable in the material. Get precise hex codes (not "medium
   blue-ish").
2. **Group by family**: all blues together, all grays together, etc.
3. **Assign semantic role** based on where they appear:

| If it appears in... | Likely role |
|---|---|
| CTA buttons, primary links | `primary` |
| General page backgrounds | `surface` or `background` |
| Card, modal backgrounds | `surface-elevated` or `surface-2` |
| Main text | `text-primary` or `foreground` |
| Secondary text, captions | `text-muted` or `foreground-muted` |
| Subtle borders | `border` |
| Positive action (success) | `success` |
| Warning | `warning` |
| Error, destructive | `error` or `destructive` |
| Decorative, hover, prominent secondary | `accent` |

4. **Detect scale** if present. Many design systems use numeric scales (50-900):
   - 50-100: very light tints (backgrounds)
   - 200-300: light (borders, subtle backgrounds)
   - 400-600: medium (the "real" color sits here)
   - 700-900: dark (text on light, dark hover)

   If you see 3+ tints of the same color used systematically, there's probably a scale.

### Suggested naming convention

```
primary, secondary, accent              ← main roles
surface, surface-elevated, surface-sunken ← backgrounds by elevation
text-primary, text-muted, text-disabled ← text by hierarchy
border, border-strong                   ← borders
success, warning, error, info           ← feedback semantics
```

---

## Typography tokens

### Identify family

In order of reliability:
1. **If CSS is accessible** and you see `font-family: "Inter", sans-serif` → certainty ✅
2. **If you visually recognize the font** (Inter, Geist, IBM Plex, Söhne, Helvetica, etc.) →
   ⚠️ medium confidence. Say "looks like Inter" not "it is Inter".
3. **If you don't recognize it**, describe it: "geometric sans with open apertures and
   straight terminals"

### Infer scale

Look for the **most common typographic measurements** in the material:

```
Display:  48-72px  (heroes, hero titles)
H1:       32-48px
H2:       24-32px
H3:       18-24px
H4:       16-20px
Body:     14-16px
Body-sm:  12-14px
Caption:  11-12px
```

Not all systems have all levels. Report only those you observed.

### Weights

List the detected weights: 300, 400, 500, 600, 700, 800. If you only saw body at 400 and
headings at 600, that's what you report. **Don't assume the system has 300-900 complete if
you only saw 2.**

### Tracking and line-height

Watch for these details that separate professional designs from amateur ones:
- **Negative tracking on large headings** (-0.02em or -0.03em) → typographic care signal
- **Generous line-height on body** (1.5-1.7) → reading-oriented design
- **Tight line-height on headings** (1.1-1.2) → typographically-present design

---

## Spacing tokens

### Infer base unit

Look at distances between elements. If most are multiples of 8 (8, 16, 24, 32), the unit is
8px. If you see many multiples of 4 (4, 8, 12, 16, 20), the unit is 4px.

**Common base units:**
- **4px** → fine-grained systems, more granularity (Tailwind default: 0.25rem = 4px)
- **8px** → classic standard, balance between flexibility and consistency
- **16px** → low-granularity systems, usually indicates simple marketing design

### Observable scale

Report the multiples you saw, not the entire possible scale:

```
Saw: 4, 8, 12, 16, 24, 32, 48, 64
Didn't see intermediate multiples — the system seems to skip (no 20, 28, 40)
```

### Suggested naming

```
space-0   = 0
space-1   = 4px
space-2   = 8px
space-3   = 12px
space-4   = 16px
space-6   = 24px
space-8   = 32px
space-12  = 48px
space-16  = 64px
```

---

## Radius tokens (border-radius)

### Common values to look for

```
none:    0
sm:      4px
md:      8px
lg:      12px
xl:      16px
2xl:     24px
full:    9999px (pill/circular)
```

### Typical patterns

- **Single system**: everything uses the same radius (e.g., all 8px)
- **Tiered system**: smaller buttons than cards (e.g., btn 6px, card 12px)
- **Deliberate mixed system**: some components with radius, others square (intentional design
  signal)

---

## Shadow tokens

If visible shadows exist, report them. Typical system:

```
shadow-sm:    minimal shadow, barely perceptible
shadow-md:    clear but visible shadow, for cards
shadow-lg:    strong shadow, for modals or elevated elements
shadow-xl:    very strong, for prominent overlays
```

If you can infer exact CSS values (rare from an image), report them. If not, describe
qualitative intensity.

---

## Output: design-tokens.json

When you extract concrete tokens, also generate a structured JSON file so it's parseable by
design-systems tooling (Style Dictionary, Figma Variables, Tokens Studio, Penpot, Supernova,
Framer, etc.).

**Format: W3C Design Tokens Community Group (DTCG)** — the format reached its first stable
version in October 2025 and is now natively supported by Style Dictionary, Figma Variables,
Tokens Studio and most major design-system tooling. Use the canonical `$value` / `$type`
form:

```json
{
  "color": {
    "primary":      { "$value": "#3B82F6", "$type": "color", "$description": "Primary action — CTA buttons, links",   "$extensions": { "anydesign": { "confidence": "high" } } },
    "surface":      { "$value": "#FFFFFF", "$type": "color", "$description": "Base background",                       "$extensions": { "anydesign": { "confidence": "high" } } },
    "text": {
      "primary":    { "$value": "#111827", "$type": "color", "$description": "Body text",                             "$extensions": { "anydesign": { "confidence": "high" } } },
      "muted":      { "$value": "#6B7280", "$type": "color", "$description": "Captions, metadata",                    "$extensions": { "anydesign": { "confidence": "medium" } } }
    }
  },
  "typography": {
    "font-family": {
      "sans":       { "$value": "Inter, system-ui, sans-serif", "$type": "fontFamily",                                "$extensions": { "anydesign": { "confidence": "medium" } } }
    },
    "font-size": {
      "body":       { "$value": "16px", "$type": "dimension",                                                          "$extensions": { "anydesign": { "confidence": "high" } } },
      "h1":         { "$value": "48px", "$type": "dimension",                                                          "$extensions": { "anydesign": { "confidence": "high" } } }
    }
  },
  "spacing": {
    "1":            { "$value": "4px",  "$type": "dimension" },
    "2":            { "$value": "8px",  "$type": "dimension" },
    "3":            { "$value": "12px", "$type": "dimension" },
    "4":            { "$value": "16px", "$type": "dimension" },
    "6":            { "$value": "24px", "$type": "dimension" },
    "8":            { "$value": "32px", "$type": "dimension" },
    "12":           { "$value": "48px", "$type": "dimension" },
    "16":           { "$value": "64px", "$type": "dimension" }
  },
  "radius": {
    "sm":           { "$value": "6px",  "$type": "dimension", "$extensions": { "anydesign": { "confidence": "high"   } } },
    "md":           { "$value": "12px", "$type": "dimension", "$extensions": { "anydesign": { "confidence": "medium" } } }
  },
  "$extensions": {
    "anydesign": {
      "source":      "https://example.com",
      "captured_at": "2026-05-18",
      "method":      "html + css-vars-extraction",
      "spec":        "W3C Design Tokens Community Group 2025.10"
    }
  }
}
```

### Notes on the DTCG conventions used

- **`$value`** holds the concrete value. **`$type`** is the canonical type
  (`color`, `dimension`, `fontFamily`, `fontWeight`, `duration`, `shadow`, etc.).
- **`$description`** is free-form prose — use it for the semantic role.
- **`$extensions.anydesign.confidence`** is where the skill-specific metadata lives. DTCG
  reserves `$extensions` for tool-specific data, so this is the legal place to keep the
  confidence marker without breaking spec compliance.
- File-level metadata (source URL, capture method, capture date) goes in the **root-level
  `$extensions.anydesign`** object.

### Legacy format (still valid)

The earlier `{ "value": ..., "confidence": ... }` shape without `$` prefixes is not DTCG
compliant but is still readable by some custom tooling. If a user explicitly requests the
legacy shape — emit it. Otherwise, default to DTCG.

### When to emit this file

**Only generate `design-tokens.json` if you extracted real tokens.** If the source was an
ambiguous image and you couldn't extract hex codes with confidence, don't invent the JSON —
stick to `design.md`.

### Color extraction helper for images

When the source is a local image and you need pixel-precise hex codes (rather than vision
approximation), run:

```bash
python scripts/extract_colors.py <image-path> --top 8
```

It returns the top-N dominant colors with their hex codes and area percentages, computed via
Pillow's quantization. Use the output to ground the color tokens in real pixel data and avoid
inventing values.

---

## Golden rule

> A short and honest system is better than a long and invented one.

If you only identified 4 colors with confidence, report 4 colors. Don't fill to 10 "because
there usually are 10". The consumer of the `design.md` trusts what you document.
rea-analysis-optimizationmorluto/reaOptimize REA providers and shared analysis workflows for agent context, latency, CPU, and memory while preserving correctness and Evidence. Use for analysis bottlenecks and cache or query design; CI/test-suite optimization belongs to rea-ci-tests-audit.Stars: 472kUpdated Claude CodeCodexVerifiedInstall
rea-ci-tests-auditmorluto/reaAudit REA CI and test suites for correctness, fixture drift, false passes, and measured performance; use for suite or workflow investigations and efficiency changes, rather than ordinary feature testing or a single test failure.Stars: 472kUpdated Claude CodeCodexVerifiedInstall
rea-tool-designmorluto/reaAudit, investigate, design, or change REA investigation tools, CLI/MCP contracts, provider capabilities, and Evidence semantics.Stars: 472kUpdated Claude CodeCodexVerifiedInstall
ponytail-auditDietrichGebert/ponytailQuality audit of the whole repo: bugs, security, real load, missing tests, speed, and what to delete. Most important first.Stars: 194kUpdated Claude CodeCodexVerifiedInstall