Visual Hierarchy Refactoring

Take a real screen or component and refactor it so the eye lands on the right things in the right order. The deliverable is refactored code plus a before/after hierarchy audit — not design commentary. Governing principles: hierarchy is a function of size, weight, and lightness contrast — color is the last resort, and the fastest fix is de-emphasis: quiet the noise before amplifying the signal.

When to use / handoffs

• Use when a specific screen is cluttered, flat, unscannable, or "looks off" despite sane tokens.
• No consistent tokens exist at all (every fix would be a new magic number) → skills/frontend-design/design-foundation first.
• The problem is structural (columns misaligned, breaks on mobile) → skills/frontend-design/layout-system.
• Font sizes themselves are chaotic across the app → skills/frontend-design/typography-system.
• Palette is the problem (clashing hues, contrast failures everywhere) → skills/frontend-design/color-system.

Step 1 — Inspect the target

1. Get the actual component. Read the file(s) for the screen the user named; if none named, ask for the one screen that matters most (this is the only intake question that blocks). Render or screenshot it if the environment allows.
2. Read the available tokens (spacing scale, text sizes, grey ramp) — every fix must use them.
3. Detect hierarchy smells in the code, with counts:
   • Emphasis inflation: how many elements per screen are bold, colored, large, or bordered? (>3 competing emphases = nothing is emphasized)
   • Cramped spacing: padding/margins ≤8px between unrelated groups
   • True greys / pure black: #808080-family, #000, text-gray-500 on colored surfaces
   • Uniform sizing: title, body, and metadata within a step of each other
   • Even spacing everywhere: identical gap inside and between groups (kills Gestalt grouping)
   • Border overuse: boxes-in-boxes where spacing would separate

Step 2 — Build the hierarchy audit

For the target screen, list every visible element and assign:

• Intended priority: P1 (the one thing this screen exists for — one, maybe two elements), P2 (supporting), P3 (metadata/chrome).
• Current visual weight: loud / medium / quiet, judged from size, weight, color, and surroundings.

Every mismatch is a work item. If you can't name the single P1, ask the user one question: "What should a user do or see first on this screen?" — then proceed.

Step 3 — Refactor, in this order

Apply changes strictly in this sequence — each pass often removes the need for the next:

1. Whitespace first. Start with too much and pull back. Double the spacing between unrelated groups; keep related items close (label 8px from its input; 24px+ between form groups). If a standard margin is 16px, try 32–48px between sections. Spacing must come from the token scale.
2. De-emphasize P2/P3 before touching P1: drop bold from labels, lighten metadata to a mid-grey step (gray-500/600), remove borders/backgrounds that box in secondary content, shrink or icon-ify tertiary actions. Secondary buttons go outline/ghost; destructive actions become quiet until confirmed.
3. Then amplify P1 — usually needs only one move once the noise is gone: one size step up, or weight 600→700, or the darkest text step, or the single saturated CTA. Not all four.
4. Contrast ladder via lightness, not hue: P1 text = darkest grey step, P2 = gray-600/700, P3 = gray-500. Tinted greys only (matching brand temperature) — never #808080/#000. Reserve saturated color for interactive elements and status.
5. Group with Gestalt rules: proximity (spacing shows relationships — the gap between groups ≥2× the gap within), similarity (identical roles look identical), common region (a card/background only when spacing can't do the job). Full principle set with CSS: references/gestalt-principles.md.
6. Size ratios: adjacent hierarchy levels differ by ~1.25–1.5×; P1 vs body ~1.5–2×. If two levels are within 10% of each other, merge them.

Canonical before/after (the shape of most fixes)

/* BEFORE: cramped, uniform, grey-on-grey */
.card { padding: 8px; }
.card h2 { font-size: 16px; font-weight: 400; color: #666; margin-bottom: 4px; }
.card p  { font-size: 14px; color: #999; line-height: 1.3; }

/* AFTER: room, clear ladder, tinted greys */
.card { padding: var(--space-6); }
.card h2 { font-size: var(--text-xl); font-weight: 600; color: var(--gray-900); margin-bottom: var(--space-2); }
.card p  { font-size: var(--text-base); color: var(--gray-600); line-height: 1.6; }
.card .meta { font-size: var(--text-sm); color: var(--gray-500); }

Required output format

Deliver code plus this audit:

## Hierarchy audit — [screen name]
| Element | Intended | Was | Problem | Change |
| Submit button | P1 | quiet (same as Cancel) | competing siblings | solid brand bg; Cancel → ghost |
| Section labels | P3 | loud (bold, dark) | emphasis inflation | weight 400, gray-500 |
| Card metadata | P3 | medium | too prominent | text-sm, gray-500 |
| ... every element with a mismatch |

## Refactored code
[full component/CSS — same styling system, tokens only]

## What changed and why (≤5 bullets)
- [pass applied → effect, e.g. "Doubled inter-group spacing (space-4 → space-8): groups now scannable"]

## Scan test
First / second / third thing the eye hits now: [P1] → [P2] → [P2]. Matches intent: [yes / remaining gap]

Quality bar (check before delivering)

• Exactly one P1 per screen, and it wins the 5-second scan
• ≤3 emphasized elements per screen; everything else visibly quiet
• Gap between unrelated groups ≥2× gap within a group; all spacing on the token scale
• ≤3 font weights in the result; bold never applied to whole paragraphs
• Text ladder uses lightness steps of tinted greys; no #808080, no #000, no new hues introduced for emphasis
• De-emphasized text still meets WCAG AA (4.5:1) — quiet ≠ illegible; P3 may go 3:1 only at large size
• Adjacent size levels differ ≥1.25×; no two levels within 10%
• Borders/boxes removed wherever spacing alone groups the content
• Every change references a token — zero new magic numbers

Integration

• Consumes: tokens from skills/frontend-design/design-foundation, type scale from skills/frontend-design/typography-system, grey ramp from skills/frontend-design/color-system, page structure from skills/frontend-design/layout-system.
• Produces: refactored components that skills/frontend-design/component-architecture generalizes into the library; contrast decisions verified by skills/frontend-design/accessibility-excellence.
• In skills/frontend-design/frontend-orchestrator paths, this runs after tokens exist (Path B step 2) or as margin-level refinement (Path C).

References

• references/gestalt-principles.md — all six Gestalt principles (proximity, similarity, figure/ground, closure, symmetry, common region) with CSS applications, plus extended before/after refactor examples. Read when a grouping problem isn't solved by proximity alone or you need more worked examples.