按约束产出可落地的设计交付物——含具体数值的调色板、组件规范、品牌指南与无障碍审计。
设计与多媒体
Design
用可量化的层级、间距、字号、配色与版式规则,绘制并诊断视觉作品。
它能做什么
围绕八条核心规则(层级排序、间距倍率、字号下限、配色对比、对齐线、先灰度设计、动效时长等),处理界面、落地页、PPT、海报、社媒图、HTML 邮件、图表与 PDF 文档。交付前按顺序运行输出闸门(灰度、眯眼、对齐线数量、间距审计、对比度、单一主视觉),并通过症状-成因对照诊断"看着不对劲"的稿件。支持 Web、iOS、Android 与印刷四种交付,默认 WCAG AA 对比度,可切换 AAA,并遵循各平台触控目标与单位规范。
什么时候用它
- 界面看着拥挤或扁平,却说不出哪里出了问题
- 在没有设计系统的情况下挑选字体并搭建配色
- 把已有页面适配到深色模式或移动端断点
- 交付前审查自动生成的 HTML 或 CSS
技能文档
User preferences and memory live in ~/Clawic/data/design/ (see setup.md on first use, memory-template.md for the file format). If you have data at an old location (~/design/ or ~/clawic/design/), move it to ~/Clawic/data/design/.
Configuration
User-dependent variables. Defaults apply until the user states a preference; store them in ~/Clawic/data/design/config.yaml.
| Variable | Type | Default | Effect |
|---|---|---|---|
| type_scale_ratio | number (1.125-1.5) | 1.25 | Regenerates the size ladder in Core Rule 4 (1.25 → 16/20/25/31/39); 1.333 for editorial and marketing pages |
| base_unit | 4 | 8 (px) | 8 | Every spacing value becomes a multiple of this; 4 for dense pro tools, 8 everywhere else |
| accessibility_target | AA | AAA | AA | AA floors are 4.5:1 body and 3:1 large text; AAA raises them to 7:1 and 4.5:1 across all checks |
| platform | web | ios | android | print | web | Switches units (px/pt/dp), touch-target minimums (44pt HIG / 48dp Material), and platform conventions in ui-screens.md and documents.md |
| brand_file | path | none | Palette, typefaces, and logo rules read from ~/Clawic/data/design/brand.md; overrides the derivation procedures in palettes.md and fonts.md |
Preference areas to record as the user reveals them:
- output medium — HTML/CSS vs image render vs written spec; framework in use (Tailwind, vanilla CSS) — affects what the deliverable looks like
- aesthetic register — minimal, dense, editorial, playful — shifts spacing generosity, type choices, and decoration tolerance
- motion posture — full motion vs reduced by default — gates Core Rule 8 and all transition suggestions
- language and script — RTL layouts mirror scan order; CJK needs looser line-height and no italic emphasis — affects
layout.mdandfonts.mdguidance - critique style — ranked punch-list vs annotated walkthrough — affects how
critique.mdfindings are delivered
When To Use
- Designing any visual artifact from scratch: UI screens, landing pages, slides, social graphics, posters, emails, PDFs.
- Fixing a design that "looks off" when the user cannot articulate why.
- Choosing typography, spacing, or a palette for a project that has no design system yet.
- Reviewing generated HTML/CSS or rendered images before delivery.
- Not for token pipelines and component libraries (design-system) or brand positioning (branding).
Quick Reference
| Situation | Play |
|---|---|
| Design looks cluttered | Delete decoration first, then double the gap between groups; never add boxes to "organize" (→ critique.md) |
| "Looks off," cause unknown | Run gates in order: grayscale, squint, alignment count (→ Output Gates); full symptom chains in critique.md |
| Everything looks equally important | Rank all content 1-3; style only rank 1 as prominent, demote the rest to body style |
| Picking fonts | One family, weights 400/700; add a second family only to split display vs text roles (→ fonts.md) |
| Picking colors | 60-30-10 split; one accent hue; neutrals tinted with 2-5% of the accent's saturation (→ palettes.md) |
| Text hard to read | Check measure 45-75ch, line-height 1.5, contrast 4.5:1 before blaming the typeface |
| Mobile layout | 44pt/48dp touch targets, 16px minimum body (also prevents iOS input zoom), primary actions in bottom thumb zone |
| Slide deck | One idea per slide, 24pt text floor, max 6 lines of text per slide (→ slides.md) |
| Poster, thumbnail, social post | One-glance artifact: single message, readable at 10% zoom (→ graphics.md) |
| Form, table, or dashboard | Component-level rules in ui-screens.md |
| Chart or data display | Zero-baseline bars, direct labels, one hue for magnitude (→ data-viz.md) |
| HTML email | Tables and inline CSS, 600px body, Outlook renders with Word (→ email.md) |
| Dark mode requested | Not inversion: lighten surfaces to show elevation, desaturate accents 10-20%, white text at ~87% opacity (Material) (→ dark-mode.md) |
| Contrast, CVD, motion sensitivity | Floors and non-color encodings in accessibility.md |
| Print or PDF deliverable | 300 DPI, bleed, CMYK, margins (→ documents.md) |
| Any other visual task | Apply Core Rules in order 1-7: hierarchy, then spacing, then type, then color |
Depth on demand: critique.md symptom→cause when it looks off · layout.md grids, alignment, responsive · fonts.md choosing, pairing, setting type · palettes.md color without a system · ui-screens.md forms, tables, dashboards, states · landing-pages.md marketing pages and CTAs · slides.md decks · graphics.md posters, social, thumbnails · documents.md PDFs, reports, print · email.md HTML email · data-viz.md charts · dark-mode.md theming · accessibility.md contrast, CVD, motion, focus.
Core Rules
- Hierarchy is a ranking, not a styling pass. Inventory every element, rank each 1-3 by user importance (1 = must see, 2 = supporting, 3 = metadata). Exactly one rank-1 per view. Check: a grayscale screenshot must reveal the ranking in 3 seconds.
- Spend one channel per level step. Adjacent hierarchy levels differ by one full step in ONE channel: size x1.25, weight +300 (400 to 700), or a clear lightness jump. Three small tweaks read as noise. Example: 18px/600 next to 16px/400 body is two half-steps and ranks poorly; use 20px/400 or 16px/700 instead.
- Spacing: base 8, factor 2 between grouping levels. Scale 4/8/16/24/32/48/64 (multiples of
base_unit). Between-group gap >= 2x within-group gap: 8px label-to-field, 24px field-to-field, 48px section-to-section. Proximity groups content more cheaply than any border or background box. - Type floors: 16px body, 1.5 line-height, 45-75ch measure. Bringhurst's 66ch is the target; under 45ch the eye jumps lines, over 75ch it loses the return sweep. Headings tighten to 1.1-1.3. Ratio
type_scale_ratio= 1.25 gives 16/20/25/31/39 and covers most product work; 1.333 for editorial and marketing pages. - Color: 60-30-10 with contrast floors. ~60% neutral surface, ~30% secondary, ~10% accent, and the accent marks interactive or state-bearing elements only. WCAG floors: 4.5:1 body text, 3:1 large text (24px+, or 19px bold) and UI component boundaries. Never #000 on #FFF; near-black (#111 to #1a1a1a) cuts edge vibration.
- Every edge answers to a line. Target <= 3 distinct vertical alignment lines per view; each extra line adds perceived clutter. Center only display text of 1-3 lines; left-align everything longer, because ragged-right gives the eye a fixed return point.
- Design in grayscale first. Layout, hierarchy, and spacing must work with zero color. Add color last to encode meaning (state, action, brand); color used to rescue hierarchy fails for the ~8% of men with color vision deficiency and fails the grayscale gate.
- Motion: 150-300ms for UI transitions. Ease-out on entry, ease-in on exit. Over 400ms reads as lag; under 100ms reads as a glitch. Animate only transform and opacity for smooth playback; honor reduced-motion preferences (
accessibility.md).
Hierarchy Procedure
- Content inventory before any styling: list every element the view must contain, including legal text and empty states.
- Rank each element 1-3. If two elements demand rank 1, split the view or demote one by position (top-left beats bottom-right in LTR scan order).
- Assign styles top-down from the type scale. Demote rank 3 by lightness (40-60% gray on white), not by shrinking below 13px, where legibility collapses.
- Place rank 1 in the natural entry point: top-left for documents, visual center for posters and slides.
- Verify with the Output Gates before delivering.
Spacing and Layout
- Macro before micro: set page margins and section gaps before touching padding inside components.
- Text content containers: 560-600px max width (keeps measure at 70-75ch at 16px, inside the 45-75ch range). Dashboards run full-width with 24px outer margins.
- Space belongs to what it labels: a heading sits closer to the text below it than to the section above (e.g. 8px below, 32px above). Equal gaps destroy grouping.
- Optical beats mathematical: icons, triangles, and play buttons need a 1-3px shift toward their visual weight to look centered. Grid math and column layouts:
layout.md. - Whitespace is the grouping mechanism. Filling "empty" corners with decoration deletes the structure the space was providing.
Type and Color Execution
- Pair typefaces by role contrast (display serif + text sans), never by similarity; two near-identical sans faces read as a rendering mistake. Selection tables and loading:
fonts.md. - Skip adjacent weights: 400 vs 500 is indistinguishable at body sizes. Use 400/600 or 400/700.
- ALL CAPS only for labels of 1-2 words, with +5-10% letterspacing. Never track lowercase body text; it breaks the word shapes readers scan by.
- Palette derivation: pick the accent hue, build a 9-step lightness ramp from it, then tint neutrals with 2-5% of the accent's saturation. Pure grays next to a saturated accent look dirty. Full procedure:
palettes.md. - Semantic color is reserved: red for destructive and error, green for success. A red brand needs error states distinguished by a different shade plus an icon, never by hue alone.
- One radius scale per artifact (e.g. 4/8/16). Mixed radii on sibling elements is the fastest tell of unreviewed generated UI.
Output Gates
Run before delivering any visual artifact:
- Grayscale test: desaturated, is the 1-3 ranking still obvious in 3 seconds?
- Squint test: blurred, do exactly the title and the primary action pop?
- Alignment count: <= 3 vertical alignment lines, and every element sits on one?
- Spacing audit: every gap a multiple of the base unit, between-group >= 2x within-group?
- Contrast check: every text/background pair meets its floor (4.5:1 body, 3:1 large/UI)?
- Single rank-1: exactly one primary action or focal point per view?
Traps
| Trap | Why it fails | Do instead |
|---|---|---|
| Boxes and borders to separate content | Each border is added noise; proximity groups for free | Remove the box, double the between-group gap |
| Shrinking text to fit the space | Below 13px legibility collapses; the content problem remains | Cut content, paginate, or widen the container |
| Centering long text | No fixed return point for the eye, each line restart costs a search | Left-align; center only 1-3 line display text |
| Emphasizing with bold + bigger + colored at once | Burns all channels on one level; nothing is left to mark the true top rank | One channel per level step (→ Core Rule 2) |
| Color as the only state signal | Fails grayscale, print, and ~8% of male users | Encode with icon, weight, or position first; color reinforces |
| Pure gray neutrals beside a saturated accent | Grays pick up a dirty, mismatched cast by contrast | Tint neutrals with 2-5% of the accent hue |
| Filling every empty area | Whitespace was doing the grouping; filling it flattens hierarchy | Leave it; if it feels empty, the content ranking is wrong, not the space |
| Restyling before ranking | Polishing elements in place cements a flat hierarchy | Content inventory and 1-3 ranking first (→ Hierarchy Procedure) |
| Designing only the happy state | Empty, loading, error, and overflow states ship broken | Design the four non-happy states before polishing the happy one (ui-screens.md) |
| Checking one theme only | A palette tuned in light mode breaks contrast in dark | Run the Output Gates once per theme (dark-mode.md) |
Where Experts Disagree
- Pure black (#000) dark backgrounds. OLED-first designers use #000 for power and contrast; Material argues #121212 reduces halation and lets shadows exist. Default #121212; go #000 only when the artifact targets OLED media consumption (video, photo feeds).
- Grid math vs optical judgment. Systematists snap everything to the 8px grid; typographers override math wherever the eye disagrees (icon centering, display-type spacing). Default: grid for layout, optical corrections allowed at component level — never the reverse.
- System fonts vs webfonts. System stacks cost zero load time and feel native; a webfont earns its bytes only when type IS the brand voice (marketing, editorial). Product UI defaults to a system stack or one variable font.
- Minimal vs dense. Consumer surfaces reward whitespace; daily-use pro tools reward density (more data per scroll,
base_unit4). The boundary is usage frequency, not taste: daily expert use → dense wins.
Related Skills
More Clawic skills, get them at https://clawic.com/skills/design (install if the user confirms):
ui— interface component patterns, interaction states, platform conventionstypography— deep type work: optical sizes, variable weight axes, print settingscolor— full color systems, tokens, and cross-surface color validationdesign-system— scaling one-off decisions into tokens and component librariesaccessibility-audit— full WCAG audit with screen-reader passes and remediation report
Feedback
- If useful, star it: https://clawic.com/skills/design
- Latest version: https://clawic.com/skills/design
Part of Clawic, the verified skill library. Get this skill: https://clawic.com/skills/design.
常见问题
- 这个技能只做评审,还是也能产出设计?
- 两者都做。它按八条核心规则生成视觉作品,并在交付前依次运行输出闸门;同时也通过症状-成因链评审已有稿件,给出可执行的修改建议。
- 覆盖哪些产出形式与平台?
- 界面、落地页、PPT、海报、社媒图、HTML 邮件、图表,以及印刷或 PDF 文档;并可在 Web、iOS、Android、印刷四种平台之间切换单位与触控目标最小值(44pt HIG / 48dp Material)。
- 无障碍方面按什么标准执行?
- 默认 WCAG AA(正文 4.5:1,大字与 UI 边界 3:1),可切到 AAA(正文 7:1,全部 4.5:1);同时显式处理色觉缺陷(不在弱视场景下丢失层级)与 prefers-reduced-motion。
相关技能
Build polished, conversion-aware frontends with strong visual taste, clear hierarchy, and production-grade HTML/CSS/JS. Landing pages, dashboards, components...
从需求收集到样图交付的营销宣传册设计流程,版式确认后才进入样机生成阶段。
用一段文字描述生成可点击的 HTML 原型——落地页、移动应用、仪表板、设计系统与线框图。
用 21 种版式与 22 种视觉风格自由组合生成信息图,并基于内容类型推荐搭配方案。
Use when users need visual direction, interface hierarchy, layout decisions, design specifications, or prototypes before implementing a Web or mini program UI.
Iván 的更多技能
浏览全部技能执行 Git 操作(提交、分支、合并、变基、冲突解决与恢复)时强制套用安全规则。
围绕 CSS 机制排查问题并编写组件样式表,而不是凭感觉试错。
以系统方式规划并执行自学:从出口测试倒推课程,加入间隔复习与刻意练习,产出可验证的迁移证据。
针对你的 Azure 订阅,做架构设计、故障排查、安全加固与成本优化
按配置的 JDK 版本诊断 Java 与 JVM 问题(从 NPE 到容器 OOM),给出可直接套用的代码与配置。
在 EC2、Lambda、RDS、VPC、IAM 等核心服务上做架构、排查、加固与成本优化,每次建议都给出月度开销与故障域。