Communitygithub.com

dashboard-ui-design

Shows an agent how to art-direct and build "dashboard ui design" so the concrete output is a landing page, homepage, or portfolio. The procedure moves through Dashboard UI design, Workflow, Non-negotiables. Use it when a new assignment calls for this craft.

dashboard-ui-design 是什麼?

dashboard-ui-design is a Claude Code agent skill that shows an agent how to art-direct and build "dashboard ui design" so the concrete output is a landing page, homepage, or portfolio. The procedure moves through Dashboard UI design, Workflow, Non-negotiables. Use it when a new assignment calls for this craft.

相容平台~Claude Code~Codex CLI~Cursor
npx skills add https://github.com/YehudaBriskman/fUdesIgn/tree/HEAD/plugins/dashboard-ui/skills/dashboard-ui-design

在你喜歡的 AI 中提問

開啟一個已預先載入此 Agent Skill 的新對話。

說明文件


name: dashboard-ui-design description: Design, build or review data-heavy dashboard UIs (analytics, finance/money, valuation, SaaS admin, ops/monitoring) with researched rules and concrete tokens. Use when creating or restyling a dashboard, KPI cards, data tables, charts, filters, tabs, popovers/dialogs, loading and empty states, number and money formatting, colour and dark mode, soft/rounded styling, motion and progressive or interactive UI, or when asked to critique, audit or score a dashboard's design. Trigger even if the word "dashboard" is not used: "admin panel", "analytics page", "metrics view", "stats cards", "portfolio screen", "make this data page look better".

Dashboard UI design

A research-backed playbook for dashboards: 85 pattern files, each with rules, anti-patterns, good and bad references, accessibility, tokens, an implementation sketch and a checklist. This file gives the workflow, the non-negotiable rules and the quick decision tables. Open the pattern files for depth.

  • Pattern index: references/INDEX.md lists every pattern with its one-line rule.
  • Pattern files: references/patterns/<prefix>-<name>.md. The prefixes are found (foundations), color and style (visual style), data (data display), chart, comp (components), motion, prog (progressive loading), and ref (product teardowns, anti-patterns, review rubric).
  • Starter tokens: assets/tokens.css, with light and dark themes. Every value in it comes from the pattern files. It is the canonical source of the shared role tokens: the other plugins' stylesheets carry the roles they read as zero-specificity defaults with these values (checked by tools/check.mjs), so each works alone and this file overrides them in any load order.

Workflow

Follow these steps in order. Each one names the files to open.

  1. Classify the view and the domain.
    • Pick the product profile (finance, analytics, ops, consumer…) for density, theme, motion budget and palette (found-domain-profiles.md), and skim the matching teardown: ref-fintech-money-dashboards.md, ref-investing-valuation-tools.md, ref-devops-dashboards.md or ref-product-analytics-dashboards.md.
    • Decide whether it is operational (live, glanceable), analytical (explore, compare) or strategic (periodic, summary). This sets the refresh rate, density, interactivity and time scale. Never mix them on one screen. See found-dashboard-types.md.
    • Write down the user's first question. The answer to it becomes the single focal point (found-visual-hierarchy.md).
  2. Structure.
    • Overview → filter → details on demand (found-overview-filter-details.md).
    • Use the inverted KPI pyramid: the answer top-leading, trends in the middle, detail at the bottom (found-scanning-kpi-pyramid.md).
    • Use a 12-column grid and a 4px spacing scale (found-layout-grid-spacing.md). Reflow by priority, with container queries per widget (found-responsive-reflow.md).
    • Disclose progressively, at most 2 levels deep (found-progressive-disclosure.md).
    • Design against worst-case data from the start: long names, huge or negative numbers, empties, 1,000 rows (found-worst-case-data.md).
  3. Pick containers and data display.
    • Choose the container: table for many comparable records, list for find-and-open, cards for a few heterogeneous summaries (data-cards-vs-tables-vs-lists.md).
    • For KPI cards, numbers, deltas, tables, timestamps and units, see the data-* files. Verdicts, ratings and scores: data-verdict-rating-badges.md.
    • Model inputs and "why this number": comp-assumption-editor-panel.md and found-methodology-explainability.md. Labels, errors and empty-state words: found-microcopy-for-data.md.
  4. Pick charts by the question, never by looks (chart-choose-by-question.md). Then apply the matching chart-* file, plus chart-honesty-chartjunk.md and chart-accessibility.md every time.
    • Finance: chart-financial-price-value.md (price vs value, waterfall, football field), chart-scenario-comparison.md (bear/base/bull, weights), chart-tornado-sensitivity.md (tornado, sensitivity grid, distributions).
    • Canvas or WebGL charts: chart-canvas-accessibility.md.
  5. Colour and style.
    • Build in neutrals first, then add colour only where it means something (color-minimal-usage.md).
    • Use three token layers: ramps → roles → components (color-system-architecture.md).
    • Keep status colours and money colours separate (color-semantic-status.md).
    • Contrast: color-contrast-rules.md. Dark mode: color-dark-mode.md. One theme per page, never mixed light and dark regions (style-theme-consistency.md).
    • For a soft, rounded look, see style-radius-scale.md, style-shadows-and-borders.md and style-surfaces-and-layering.md.
  6. Components and interaction.
    • Buttons, tabs, overlays, filters, date ranges, ⌘K, inline edit, sliders, app shell, alerts and keyboard: see the comp-* files.
    • For which overlay to use, start at comp-overlay-choice.md. First-run notices and disclaimers must not cover content (comp-onboarding-and-notices.md).
    • Link every entity a page names to its own detail view, so users go detail → detail without the sidebar; add hover/focus previews (300–500 ms hover delay, immediate on keyboard focus, 250 ms close grace, content chosen by context, touch and keyboard alternatives) for the quick follow-up question (comp-entity-links-and-hover-previews.md).
  7. Motion and progressive UI.
    • Motion tokens: motion-principles-and-tokens.md. Reduced motion: motion-reduced-motion.md.
    • Loading: prog-loading-states.md and prog-stale-while-revalidate.md.
    • Interactive exploration: motion-linked-exploration.md and motion-what-if-live-recompute.md.
    • Performance budgets: prog-perceived-performance-thresholds.md.
    • Drag, sheets and flicks: motion-drag-and-spring-physics.md; for general product motion use the motion-ui plugin.
  8. Review before calling it done.
    • If the design-review plugin is installed, run its static lint and rendered gate first; both must pass at zero.
    • Run ref-dashboard-anti-patterns.md (10 failures, each mapped to rubric IDs) and check for the generated-dashboard tells in ref-ai-default-tells.md.
    • Score ref-design-review-rubric.md: 39 checks in 8 sections, 0–2 each, scored per page type. Ship at 80% or more with no must-check at 0 on any page type.
    • Report the score per page type and each failed check with its ID, evidence and fix.

Non-negotiables

Break one of these only with a written reason.

Hierarchy and layout

  1. Each view has one primary focal point that answers the first question; size, contrast and position are spent there first.
  2. Every number has context: a unit, a period, and a comparison, target or trend. A bare number is not a KPI.
  3. Group with space before lines or boxes. Gaps inside a group are smaller than gaps between groups. Use cards for self-contained units, not as default wrapping.
  4. Each KPI row has at most about 4–6 cards, all with the same anatomy and size. More KPIs means you need a table or a second view.

Numbers and money

  1. Use one locale-aware formatting layer (Intl.NumberFormat), with a true minus sign (U+2212, normalised via formatToParts) and the same precision within a column or row. Keep the exact value one hover or focus away.
  2. Use tabular lining figures for every number. Right-align numeric columns and their headers.
  3. Show a delta as sign, then magnitude, then basis, in text ("+4.2% vs last 30 days"). Arrow and colour are redundant cues. Colour encodes desirability per metric (up can be bad), never direction alone. Write "pp" for percentage points.
  4. Never show missing as zero. Use a muted em dash with visually hidden "No data".

Colour

  1. Build in neutrals first, then add colour only to encode meaning. Aim for 10% or less of the screen in saturated colour.
  2. Text contrast is 4.5:1 or more (3:1 for large text). Component boundaries, focus rings, icons and chart marks are 3:1 or more. Colour is never the only cue.
  3. Money positive/negative tokens are separate from success/danger. Red-up markets exist (CN, JP, TW, KR), so offer red-up, green-up and colour-blind-safe blue/orange settings.
  4. Categorical palettes use 6 colours by preference, 8 as a hard cap, then "Other" or small multiples. Use sequential for ordered data and diverging only when the midpoint means something.

Charts

  1. Choose by the comparison (change, rank, part-to-whole, distribution, deviation, correlation, flow, magnitude). Bars and areas start at zero; lines may crop. No 3D, no dual axes for general readers, and no pies with more than 5 slices. Replace gauges with bullet graphs.
  2. Label directly at line ends and on marks. Use a legend only when labels can't fit. Tooltips are detail on demand, and hover, focus and touch all trigger them.
  3. Every chart has a takeaway title, a one-line text summary and a table view, and uses non-colour encodings.
  4. Show uncertainty as labelled ranges ("80% of simulated outcomes"), never a single confident line.

Components

  1. One primary button per region. Targets are 24×24 CSS px or more (44 px on touch). A busy button keeps its size and label and uses aria-disabled, not disabled.
  2. Tabs switch content, segmented controls switch the mode of one view, and pills filter. View state (tab, filters, range, sort) lives in the URL.
  3. Use the least intrusive overlay. Use a modal <dialog> only for a decision needed now. Errors appear inline next to their cause and persist; toasts are a last resort and never the only record.
  4. Date ranges: lead with presets, show the resolved absolute dates and timezone, and make the comparison period explicit and length-matched.
  5. Everything works by keyboard with a visible focus ring. Composite widgets are one Tab stop with arrow keys inside. Single-key shortcuts can be turned off (WCAG 2.1.4).

Motion and loading

  1. Motion only explains a change. Most UI motion is 300 ms or less; use the token durations of 100/150/240/400 ms. Don't animate actions users repeat dozens of times a day.
  2. Under prefers-reduced-motion: reduce, remove movement but keep opacity and colour feedback. Never use a global 0.01ms reset.
  3. Loading depends on the wait:
    • under 1 s, nothing (delay indicators about 200 ms);
    • 1–10 s on first load, a layout-accurate skeleton;
    • one panel refreshing, in-place;
    • over 10 s, determinate progress with cancel.
  4. Once a panel has shown data, never blank it again. Keep the stale data visible with "Updating…" and a last-updated time (stale-while-revalidate).
  5. Budgets: visual acknowledgement under 100 ms, INP 200 ms or less at p75, result or clear progress under 1 s. What-if sliders recompute while dragging, or show a provisional value and commit on release.
  6. Every region has explicit loading, empty, zero-results, error and partial states, each scoped to the smallest region that has it.

Trust

  1. State data freshness ("Updated 4 min ago", with the exact time and timezone on hover) and the source and definition next to the numbers they qualify.
  2. No celebratory decoration on money actions: a 2024 Massachusetts consent order requires Robinhood to permanently cease confetti and other celebratory imagery tied to trading frequency. Keep judgement glyphs and verdicts traceable to the raw numbers in one step.

Decision tables

Container (data-cards-vs-tables-vs-lists.md)

TaskUse
Compare many records across attributesTable: sticky header and first column, sortable, 20–50 rows per page
Find and open one itemList rows: leading visual, label, one meta line, one trailing value
A few heterogeneous summariesCards

Overlay (comp-overlay-choice.md)

NeedUse
Blocking decision nowModal <dialog>
Contextual detail or small form, page still usablePopover (popover API, Baseline widely available since October 2026; anchor positioning with an @supports fallback to fixed placement)
Large secondary task beside the contextDrawer or sheet
Low-stakes, recoverable confirmationToast, persistent if it holds an action
Name or hint of an icon or termTooltip, never essential content (popover="hint" and interestfor are not Baseline: feature-detect)

Loading (prog-loading-states.md)

Expected waitTreatment
Under 1 sNothing; delay any indicator 200 ms, keep it visible at least 400 ms once shown
1–10 s, first loadLayout-accurate skeleton
One panel refreshingKeep the data, show "Updating…" (prog-stale-while-revalidate.md)
Over 10 sDeterminate progress with cancel, can continue in the background

Chart by question (chart-choose-by-question.md)

QuestionChart
Change over timeLine (area if the total matters)
RankSorted bars, horizontal for long labels
Part to wholeStacked bar, or a donut with 2–5 parts
DistributionHistogram or strip plot
UncertaintyFan or band chart, labelled interval
Deviation from a referenceDiverging bars
CorrelationScatter
Bridge between two valuesWaterfall
Methods compared to priceFootball field
Value vs targetBullet graph
Trend next to a numberSparkline

Token defaults

Full set in assets/tokens.css.

GroupDefaults
Spacing (4px base)4, 8, 12, 16, 24, 32, 48, 64
Radiusxs 4 · sm 6 · md 8 · lg 12 (cards) · xl 16 (dialogs) · full. Nested radius: inner = max(2px, outer − padding − border)
Typehero 2rem/600 · kpi 1.875rem/600 · tracking −0.02em on hero and KPI · title 1.25rem · heading 1rem/600 · body 0.875rem · meta 0.75rem · prose 65ch or less · tabular-nums
Motion0 instant · 100 micro · 150 short · 240 medium · 400 long · 600–900 staged data. Standard ease cubic-bezier(0.2,0,0.38,0.9), emphasized cubic-bezier(0.2,0,0,1). Stagger 20–40 ms, 300 ms max total
Elevation4 levels: flat, raised, popover, dialog. Layered, hue-tinted shadows; borders for in-flow surfaces
Surfacescanvas, surface, surface-2, surface-3, sunken; plus on-accent, scrim, hover, zebra, chart grid/axis/ink roles. Density: [data-density="compact"]. Dark mode uses dark grey (not black), lighter means higher, desaturated accents
KPI cardpadding 16–24 · radius 12 · value 1.875rem/600 · sparkline 120×28 · 4–6 per row at 1280px or wider, 2 at 768px or wider, 1 below
Grid4 columns under 600px · 8 at 600–1055px · 12 at 1056px or wider · gutter 16 (24 at 1312px or wider)

How to apply this skill

  • When building:
    • Start from assets/tokens.css (or map its roles onto the project's existing tokens), then follow the workflow.
    • Prefer platform features: <dialog>, the popover API, anchor positioning, container queries, View Transitions (fetch first, then transition), :has(). The implementation sketch in each pattern shows the feature detection and fallback.
    • Never use the TypeScript any type in sketches or code.
  • When reviewing: collect evidence (screenshots, DOM, CSS), then run the anti-pattern catalog and the rubric. Report a table of the check, the score, the evidence and the fix, ordered by severity.
  • When the project has a design system: its tokens win, so map them onto these roles and flag the gaps instead of overriding them.
  • When rules conflict: accessibility and honesty (contrast, non-colour cues, zero baselines, states) beat aesthetics. Evidence beats taste: ref-ux-research-findings.md lists what is actually evidenced.

Individual skills in this repo

This repo contains 3 individual skills — each has its own dedicated page.

相關技能