Communitygithub.com

AlessandroFare/fluxychat

Design and improve user interfaces using the concrete rules from Refactoring UI (Wathan & Schoger) — constrained spacing/type/color/shadow scales, visual hierarchy through weight and color rather than size, and depth through emulated light. Use when building or styling any UI (web, app, dashboard, landing page, component), when picking font sizes, spacing, colors, shadows or border radius, when designing a color palette or design tokens, and whenever someone says a UI "looks off", "looks amateur", "feels cluttered/plain/unfinished", or asks to "make this look better".

fluxychat 是什么?

fluxychat is a Claude Code agent skill that design and improve user interfaces using the concrete rules from Refactoring UI (Wathan & Schoger) — constrained spacing/type/color/shadow scales, visual hierarchy through weight and color rather than size, and depth through emulated light. Use when building or styling any UI (web, app, dashboard, landing page, component), when picking font sizes, spacing, colors, shadows or border radius, when designing a color palette or design tokens, and whenever someone says a UI "looks off", "looks amateur", "feels cluttered/plain/unfinished", or asks to "make this look better".

兼容平台~Claude Code~Codex CLI~Cursor
npx skills add https://github.com/AlessandroFare/fluxychat/tree/HEAD/.agents/skills/refactoring-ui-skill-main

在你喜欢的 AI 中提问

打开一个已预加载此 Agent Skill 的新对话。

文档

Refactoring UI

Visual design is not talent. It is a small set of systems decisions made once, plus a handful of techniques for creating hierarchy. This skill is those systems and those techniques.

The single biggest cause of amateur-looking UI is picking values ad hoc — 17px here, #3B82F6 there, lighten(5%) for a hover state. Design from a scale, always.


The systems — use these, do not re-derive them

Pick from these lists. Never invent a value that isn't on one.

Spacing and sizing

Base 16px, built from factors and multiples of it:

4  8  12  16  24  32  48  64  96  128  192  256  384  512  640  768

Tight at the small end, spreading out at the large end. No two adjacent values may be closer than ~25% — that is what makes the choice obvious. A linear "multiples of 4" scale fails: it does not help you decide between 120px and 124px.

Use it for margin, padding, width, height, icon sizes, border width — everything spatial. Fix a small set of opacity values too (e.g. .05 .1 .2 .4 .6 .8) for disabled states, overlays and hover tints, rather than eyeballing a slider each time. Same logic as the other scales: decide once, reuse everywhere.

Type scale

12  14  16  18  20  24  30  36  48  60  72

Not a modular scale built from a ratio (4:5, 2:3, golden ratio). Those produce fractional pixel values that round inconsistently across browsers and are too sparse for interface density — hand-picked values, chosen for how they feel, win instead.

Units: px or rem only. Never em. em is relative to the current font size, so a .875em inside a 1.25em parent computes to 17.5px — a value not in your scale. The scale silently stops existing.

Font weight

Two weights is enough:

  • 400 or 500 — body and most UI text
  • 600 or 700 — anything emphasized

Nothing below 400 in UI. To de-emphasize, use a lighter color or smaller size — never a lighter weight.

Color

You need far more colors than a five-swatch palette generator gives you.

  • Greys: 8–10 shades. Almost all of a UI is grey — text, backgrounds, panels, borders, form controls. Three or four shades always runs out. Start at a very dark grey, not true black (true black looks unnatural).
  • Primary: 5–10 shades, one or maybe two primaries.
  • Accents: 5–10 shades each — destructive red, warning yellow, positive green, plus whatever else the product needs to distinguish (chart series, calendar events, tags). Ten colors × 5–10 shades is normal for a complex UI.

Name them 100 (lightest) → 900 (darkest), base 500.

Build the scale in this order: pick 500 first (for a primary/accent, it should be a shade that works as a button background). Then find the edges — 900 is usually your text color, 100 a background tint; an alert component uses both, so design one and read the two values off it. Then fill 700 and 300 as the perfect compromise between their neighbours, then 800 600 400 200 the same way.

Write colors as HSL, not hex. hsl(220, 95%, 34%) and hsl(220, 65%, 61%) are visibly related; #03369E and #507DD7 are not.

Never generate shades at runtime with lighten() / darken(). That is how you end up with 35 slightly different blues.

Shadows — five elevations

0 1px 3px  hsla(0,0%,0%,.2)   /* barely raised — buttons */
0 4px 6px  hsla(0,0%,0%,.2)   /* dropdowns */
0 5px 15px hsla(0,0%,0%,.2)
0 10px 24px hsla(0,0%,0%,.2)
0 15px 35px hsla(0,0%,0%,.2)  /* modals */

Choose by asking where on the z-axis does this sit?, not what shadow looks nice?. Closer to the user = more attention. Shrinking a button's shadow on :active makes it feel pressed; growing a list item's shadow when it's picked up for drag-to-reorder does the reverse — it reads as "now above its siblings" and doubles as the drag affordance itself.

These five are the default and are fine everywhere. references/techniques.md gives a refined two-part version of the same scale — parallel, but not identical values (its top step is heavier). Use it when shadows are prominent in the design; use these when they aren't. Don't mix the two in one project.

Line-height and line length

Line-height is inversely proportional to font size, and proportional to line width:

  • small text / wide columns → 1.5 to 2
  • large headlines → 1 is fine

Line length: 45–75 characters, i.e. max-width: 20em–35em. This applies to the paragraph even when the container around it is wider — mixed widths in one content area look more polished, not less.

(em is correct here — measure should scale with the text it wraps. The "never em" rule is scoped to the type scale, where em compounds through nesting. Don't "fix" this.)

Border radius

Pick one and stay consistent. Small radius = neutral. Large radius = playful. None = serious/formal. Mixing square and rounded corners in one interface always looks worse.


The procedure

1. Start with a feature, not a layout. Don't design "the app" — you cannot decide between top nav and sidebar before you know what's in the product. Design one real piece of functionality (the search form, the message composer), and let the shell emerge.

2. Detail comes later. Ignore typefaces, shadows and icons early. Work in grayscale first — it forces hierarchy to come from spacing, contrast and size rather than color. Add color once the layout works.

3. Don't over-invest in low fidelity. Sketches, wireframes and mockups are disposable — nobody can use a static picture of an app. They exist to explore ideas; abandon them once the decision is made and go build the real thing.

4. Design the smallest useful version, then build it. Work in short design→code cycles. Don't imply functionality you aren't ready to build — a comment box with an attachments zone you can't ship yet blocks the whole feature. Nice-to-haves get designed later.

5. Choose by elimination. When picking a value from a scale: guess the one you think is right, then compare it against the neighbour on each side. Two will be obviously wrong. If an outer option wins, re-run the comparison with that as the new middle.

6. Start with too much white space and remove it. Adding space until something stops looking bad gives you the minimum. Starting generous and trimming gives you the right amount. Dense UIs (dashboards) are legitimate — but as a deliberate decision, not a default.

7. Shrink the canvas. A small component designed on a 1400px artboard tends to sprawl, because the space is there to fill. Start at ~400px and design the mobile layout first, where the constraints are real — then bring it to a large screen and relax only what genuinely felt cramped. You will change less than you expect.


Hierarchy — the technique that does the most work

Everything on screen sits in a pyramid: primary, secondary, tertiary. When everything competes, the UI reads as noise. This is what makes a design look "designed" — not styling.

Size isn't everything. Leaning on font size alone gives you primary content that's too big and secondary content that's too small. Use weight and color to carry emphasis instead, and keep sizes reasonable.

Three text colors, maximum:

  • dark — primary content
  • grey — secondary content
  • lighter grey — tertiary (footnotes, copyright)

All three carry real body-size text, so all three need 4.5:1. "Lighter grey" means the lightest shade that still clears it — roughly the middle of a 9-step ramp, not the pale end. The pale shades are for disabled states and large text only.

Emphasize by de-emphasizing. When the important element won't stand out and there's nothing left to add to it, soften what competes with it instead. Fade the inactive nav items; drop the sidebar's background color so the main content sits forward.

Actions: style by hierarchy, not by semantics.

  • Primary → solid, high contrast. Usually exactly one per page.
  • Secondary → outline, or a low-contrast background.
  • Tertiary → styled like a link.

Destructive ≠ big red button. If "Delete" isn't the primary action on the page, give it tertiary treatment — then make it a big red primary button inside the confirmation dialog, where it genuinely is the primary action.

Labels are a last resort. label: value gives every piece of data equal weight. Most data identifies itself by format ($19.99, an email address) or by context. Where a label is genuinely needed, fold it into the value ("12 left in stock", not "In stock: 12"), or add it as visibly secondary content. Exception: on spec-sheet-style pages where users scan for the label, emphasize the label instead.

Balance weight against contrast. Solid icons are visually heavy and will out-shout the text beside them — soften their color to compensate. It works in reverse too: when a 1px border is too subtle in a soft color but too harsh once you darken it, keep the soft color and go to 2px. Add weight to fix low contrast; reduce contrast to fix excess weight.

Visual hierarchy ≠ document hierarchy. Semantic markup and visual weight are separate decisions. Section titles are usually labels, not headlines — an h1 at 16px is fine, and sometimes the title should be visually hidden entirely because the content speaks for itself.


Hard rules

Deviating from these produces a specific, recognizable failure.

  1. Never grey text on a colored background. Grey-on-white works because it reduces contrast; grey on color just looks dirty. White-at-reduced-opacity looks washed out and disabled, and lets patterns show through the glyphs. Hand-pick a color with the background's hue, adjusting saturation and lightness.
  2. Never em for the type scale. px or rem.
  3. Never generate shades at runtime. Define them up front.
  4. Never use a percentage width for something that shouldn't scale. Sidebars get fixed widths; the main area flexes. Elements get a max-width and only shrink when the screen is actually smaller — a login card shouldn't be wider at medium screens than at large.
  5. Never scale things proportionally across breakpoints. Large elements must shrink faster than small ones. A 2.5em headline sitting on 14px mobile body copy computes to 35px — far too big; it wants to be 20–24px there. Likewise a button's padding should get proportionally tighter as the button shrinks, not scale with its font size.
  6. Always more space around a group than within it. This is the fix for "which label belongs to which field", cramped bullet lists, and headings that look attached to the wrong paragraph. Ambiguous spacing is a functional bug, not just an ugly one.
  7. Never use color as the only signal. Add an icon, a shape, or a text cue. For charts, distinguish series by contrast (light→dark shades of one color) rather than by hue — colorblind users read lightness reliably, hue not so much.
  8. Contrast minimums: 4.5:1 for normal text. The 3:1 allowance applies only to large text, which WCAG defines as 24px regular or 18.66px bold — not 18px. Assume 4.5:1 unless the text is genuinely that large. When white-on-color fails, flip it: dark colored text on a light colored tint (see references/systems.md). Separately, 3:1 applies to non-text too (WCAG 1.4.11): if a border is the only thing identifying a control — an input outline, a checkbox edge — it needs 3:1 against its background. A hairline that merely divides content does not. Those are two different tokens, not one.
  9. Never scale an icon far from its intended size. A 16–24px icon at 48px looks chunky and detail-starved. Put it inside a colored circle instead.

References

assets/tokens.css — a complete, contrast-verified starting set of all the above as CSS custom properties: spacing, type, weights, border width, a fixed opacity scale, a 9-shade cool grey ramp, a 9-shade primary, three accent trios, five elevations, a semantic role layer (--surface, --text-primary, --action…) and a dark-mode block that overrides only those roles. Copy it in and retune the hues rather than re-deriving the scales from this prose. Reference the roles in components, not the raw ramps — that is what makes the dark mode work. Every text/surface pair is verified ≥4.5:1 in both modes and every functional border ≥3:1; the deliberate sub-threshold shades (disabled, decorative dividers, the large-text-only step) are exempt by criterion and commented inline.

Load these when the work calls for them:

  • references/systems.md — building a palette from scratch: choosing the base color, keeping saturation alive at the light and dark ends, hue rotation, warm/cool greys, and the two escape hatches for hitting contrast ratios without ugly color.
  • references/diagnose.md — symptom → fix table. Load first whenever the task is improving existing UI rather than building new.
  • references/techniques.md — depth and light simulation, two-part shadows, baseline alignment, letter-spacing, breaking out of default component shapes, and handling user-uploaded images.

Individual skills in this repo

This repo contains 10 individual skills — each has its own dedicated page.

AlessandroFare/fluxychat

Room layer on Cloudflare. Chat, presence, Yjs, and agents on one Durable Object. MIT self-host or hosted beta.

AlessandroFare/fluxychat

Use when audio already placed in a HyperFrames composition needs to be mixed: a music bed that fights a voiceover (voiceover carve), effects on a track (EQ, compressor, limiter, gate, saturation, delay, reverb, chorus, phaser, bitcrush), or automation envelopes drawn on a track's volume or any effect parameter. Don't use for sourcing or generating audio — finding BGM, SFX, or making a voiceover is `/media-use`. Don't use for clip timing or track layout, which is `/hyperframes-core`.\n

AlessandroFare/fluxychat

Use the HyperFrames CLI development loop: init, add, catalog, capture, lint, check, snapshot, compare, grade-compare, preview, play, present, beats, keyframes, single or batch render, publish, cloud, cloudrun, feedback, lambda, doctor, browser, info, upgrade, skills, compositions, docs, benchmark, telemetry, transcribe, auth, tts, and remove-background. Also use when diagnosing build or render failures. validate, inspect, and layout are deprecated aliases; use check. Covers local, HeyGen-hosted cloud, AWS Lambda, and Google Cloud Run rendering.\n

AlessandroFare/fluxychat

Install, discover, and wire registry blocks and components into HyperFrames compositions. Use when running hyperframes add or hyperframes catalog, installing one item or every block matching a tag, wiring an installed item into index.html, or working with hyperframes.json. Covers discovery, install locations, block sub-composition wiring, component snippet merging, and authoring a new block or component to contribute upstream (idea → scaffold → validate → PR).

AlessandroFare/fluxychat

Mandatory entry point: read this first for any request to make, create, edit, animate, or render a video, animation, or motion graphic, including a promo, explainer, captioned clip, title card, overlay, slideshow or interactive deck, Remotion port, or any HyperFrames HTML composition. Also use it to inspect, diagnose, validate, preview, publish, or batch-render an existing HyperFrames project. Inputs may be a website URL, GitHub PR, Figma design or URL, text or brief, existing footage, or music. It resumes project state, captures intent when applicable, selects and installs the owning workflow, and routes domain capabilities. HyperFrames is the default output framework unless the user explicitly chooses another framework for the deliverable or asks only to record a browser session.\n

AlessandroFare/fluxychat

Overlay doctrine for the embedded-captions workflow — the caption MODEL (drop / rail / embed) and the rule that captions are an OVERLAY composited on top of the film, never a reserved bottom band you shift content up to avoid. Load when adding captions/subtitles to a talking-head or launch video, when deciding whether a phrase should be dropped, ride the verbatim rail, or be promoted to a scarce embedded climax, when laying out a composition that will carry captions (do NOT reserve a keep-out band), or when centering a composition on the true frame center under captions. Quotes the rail+embed model from embedded-captions and constraint #13 (captions overlay, keep-out band retired) from the product-launch-video scene agent. Applies ON TOP of embedded-captions.

AlessandroFare/fluxychat

Turn a weekly changelog .md into a finished branded changelog video (square 1080, ~45-60s, Annie VO, animated brand background, mock-UI visualizations, lowkey captions). Use when the user provides a changelog/digest markdown and wants the weekly video, or says "changelog video". Self-contained — fonts, background, lexicon, and scripts ship in this skill.

AlessandroFare/fluxychat

Author or edit a custom HyperFrames composition when no specialized workflow fits, or when BRIEF.md sets flow: companion. Use for longer or multi-scene pieces, brand and sizzle reels, montages, static loops, static title cards, footage remixes, and freeform builds. Use motion-graphics instead for a short unnarrated motion-first unit, including an animated title. Route fresh creation through hyperframes before using this skill.

AlessandroFare/fluxychat

All animation knowledge for HyperFrames — atomic motion rules, multi-phase scene blueprints, scene transitions, broader motion-design techniques, AND the seven runtime adapters (GSAP default, plus Lottie, Three.js, Anime.js, CSS keyframes, Web Animations API, TypeGPU). Use for any motion or animation task: pick 2-4 rules and compose, or load a blueprint, or look up runtime-specific API (e.g. GSAP eases / Lottie player / Three.js mixer). Also covers auditing an existing composition's choreography (animation map) and 24 named text-animation effects. HyperFrames-native: single paused timeline, seek-safe, deterministic.

AlessandroFare/fluxychat

Use before UI-related work to select the smallest useful UI Skills context through the ui-skills CLI.

相关技能