agentleFS
iamcaominhtieniamcaominhtien/kanban-board-mcp/.claude/skills/canvas-design/SKILL.md

canvas-design

Skill1 starsChanged 4 days agoMITView on GitHub

Design the HTML frames of a Canvas (mockups and prototypes of screens, pages and flows) so they look like one considered product, presented as a design document (foundations, components, then screens, each board with its states side by side), then create them with docs_write(canvas_create) or change a frame with canvas_update. Use when asked to design, mock up or redesign a screen, an app or a flow on a canvas, or to explore design directions.

198 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use academy-guide, skill-share and 10 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • AnthropicSkill

    academy-guide

    Stop and check this skill before finishing any reply to a question about how to use Claude or a Claude product — it recommends matching courses, tutorials, and use cases from Claude Academy (academy.claude.com), Anthropic's learning hub. Trigger on: "how do I", "how can I", "getting started with", "what can Claude do", "teach me", "learn to use"; questions about artifacts, projects, skills, plugins, connectors, MCP; requests about rolling Claude out to a team, class, or organization; and any ask for training materials, onboarding content, or learning resources. Use it when the user is learning how to use a feature or product — not when they are mid-task and just want the task done. This skill composes with other skills: after consulting product documentation to answer how a Claude feature works, also check here for a matching course or tutorial — a docs-grounded answer and an Academy recommendation belong together. Only recommend on a strong match; never invent Academy content.

    Also in 11 projects that use this

  • ArogyaReddySkill

    skill-share

    A skill that creates new Claude skills and automatically shares them on Slack using Rube for seamless team collaboration and skill discovery.

    Also in 28 projects that use this

  • ArogyaReddySkill

    meeting-insights-analyzer

    Analyzes meeting transcripts and recordings to uncover behavioral patterns, communication insights, and actionable feedback. Identifies when you avoid conflict, use filler words, dominate conversations, or miss opportunities to listen. Perfect for professionals seeking to improve their communication and leadership skills.

    Also in 37 projects that use this

  • zephyrwang6Skill

    topic-collector

    AI热点采集工具。从Twitter/X、Product Hunt、Reddit、Hacker News、博客等采集AI相关热点内容。当用户说"开始今日选题"、"采集热点"、"看看今天有什么新闻"、"今日AI热点"时触发。聚焦领域:Vibe Coding、Claude Skill、AI知识管理、AI模型更新、AI新产品、海外热点。

    Also in 4 projects that use this

  • zephyrwang6Skill

    deepl

    Übersetzt Texte, Dokumente und XLIFF-Dateien via DeepL API. Verwende für Übersetzungen mit /deepl translate, /deepl file oder /deepl xliff.

    Also in 4 projects that use this

  • Patrick-Git-CreatorSkill

    artifacts-builder

    Suite of tools for creating elaborate, multi-component claude.ai HTML artifacts using modern frontend web technologies (React, Tailwind CSS, shadcn/ui). Use for complex artifacts requiring state management, routing, or shadcn/ui components - not for simple single-file HTML/JSX artifacts.

    Also in 51 projects that use this

  • +3 more from ArogyaReddy
  • +3 more from zephyrwang6

The file

We split this skill into the rules it gives an agent

7 rules. 0 are common, 5 are distinctive, and 2 are very specific to this project.

  • 0

    Common

    Most of the 190 edited versions we read keep these.

  • 5

    Distinctive

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Do not converge on the same choices from one brief to the next.

    line 64
  2. Distinctive:

    Every state says what happened and what to do next ("Plan for 4 tickets is waiting for your approval", "Needs you: choose a sender …

    line 72
  3. Distinctive:

    Every element earns its place: no filler sections, decorative numbers, icons or stats.

    line 74
  4. Distinctive:

    No fake phone chrome: no painted status bar ("9:41", battery), no fake keyboard.

    line 106
  5. Distinctive:

    Do the boards read as a document: pages in order, a title and purpose on each, lettered sections, states side by side with labels, a …

    line 165
  6. Very specific:

    Change to an existing frame: write the changed files to a folder and docs_write(canvas_update) with a one-sentence note on the design …

    line 191
  7. Very specific:

    Leave canvas_comments for open questions, each pinned with at (a CSS selector, or text=Add task).

    line 193

“Common” is measured: kept by at least 6 in 10 of the 190 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Designing frames for a Canvas

A frame is an open web page (scripts, CDN libraries and fonts run). Good frames come from a process, not from more CSS: context, a direction, foundations, then the boards, then a check. This one text is what an AI team with a designing role and every MCP agent are given (it is the GUIDE in server/services/canvas_design.py; the skill file in .claude/skills/canvas-design/ is generated from it and a test fails if they differ).

Beside this guide (the skill's folder; outside Claude Code, docs_read(design_guide) gives craft and the examples' path): craft.md, how a design stops looking machine-made (read it before the first frame), and two exemplar boards in examples/, foundations.png and states.png (each with its .html). The exemplars are the bar for how a design is presented (a document, states side by side, captions). They are a work tool's UI: never take their look (green, pills, hairline cards, the fonts) for another product, which gets its own.

Other design skills never set the look. A board run's prompt lists the skills installed for you, each by its exact name and the work it covers. Colour, type, layout and the signature come from this guide and craft.md, never from a skill's rule set: a palette derived by rule from one hue, a stock type scale and white cards on a tinted ground left a live run flat, in one hue. Open a skill with the Skill tool only for a craft this guide does not teach (motion, 3D, a chart) when you reach that work, or to check boards you have drawn (critique, accessibility, the handoff spec). Where one disagrees with this guide, the guide wins. The canvas speaks the product's language: board titles, captions and notes are in the language of its copy.

0. Context before anything

Design is rooted in what exists. Before the first line of code, look for the product's own screens (other frames, a ticket or page that describes it), its brand or design system, its code (components, CSS variables) and screenshots. Match its vocabulary: tone of the copy, palette, density, radius, shadows, hover and pressed states. Designing a whole product from nothing is the last resort; if you must, say what you assumed. Where you have no icon or asset, draw a plain labelled placeholder. To recreate a UI whose source you can read, build from that source (exact paddings, radii, sizes), not from memory. Do not recreate another company's distinctive proprietary UI; make an original design that does the same job.

1. Ask once, then decide

With a person there to answer (Claude Code can use AskUserQuestion), ask one round of at most 4 questions before the first frame, only if two or more result-changing decisions are open (which job the screen does first, who lands on it, what to reuse or break from, what may be added). Questions come from their material, never a stock form. Each has 2 to 4 concrete options that differ on a named axis, your pick first marked "(Recommended)". Never re-ask a settled point. When nobody can answer (a headless run, "just make it"): decide, build, and state your assumptions in one line.

2. Direction: character before pixels

Name these and keep them for the whole canvas: purpose and user, tone, the signature, and type, colour, space.

  • The signature is an element only this product would have, which shapes its main screen and still reads at thumbnail size: a budget whose month is an envelope that empties, a transit app drawn as its own line map, a habit tracker as a chain that gains a link a day. Find this product's own; never reuse an example from this guide. A 2 px tick on a progress bar is a detail, not a signature. If the name of the app could be swapped and nothing would look wrong, there is no signature yet.
  • Calm is not characterless. An app used every day carries no decoration (no hero art, no gradient wash, no ornament), and it still has a face: a type pairing chosen for this product, a colour that owns the screen, a composition, the signature. A landing page, a splash or a poster can go further: a bold display face, a ground with depth, a grid-breaking layout.
  • Explore before you build. For a new product or a new main screen, draw that screen three ways that differ on a named axis (the signature, light or dark, density, the type), at full fidelity with the same real data, on a "Directions" page. With a person to answer (AskUserQuestion, or ai_run(ask) in a board run), show them and let them choose. Otherwise judge each at thumbnail size against the tells in craft.md, keep the strongest, and say why. The rest of the canvas follows the chosen one; the choice stays chosen.
  • Do not converge on the same choices from one brief to the next. A cream ground, a serif display and terracotta, or a white app of rounded cards with one blue, are habits, not directions. Inter, Roboto or the system face as the display face is the template look, unless the product already uses it.

3. Content

  • Real copy, written for this product: real names, amounts, dates, counts, error messages. Never lorem ipsum, never filler that could describe any business. A missing fact is a visible placeholder like [YOUR PRICE].
  • Every state says what happened and what to do next ("Plan for 4 tickets is waiting for your approval", "Needs you: choose a sender address"), never blame and never a bare "Error".
  • Every element earns its place: no filler sections, decorative numbers, icons or stats. Ask before adding sections, pages or copy nobody mentioned.

4. Tokens

One tokens.css copied into every frame (or one :root block), and nothing in a frame that is not a token.

TokenRule of thumb
Neutralsink, muted, line, surface, background; tinted slightly toward the brand (never pure grey on pure white)
Colourin oklch(): one colour that owns the screen (a ground, a fill, the signature), at most two accents sharing its chroma and lightness; state colours used sparingly, on the thing they concern
Spacing4, 8, 12, 16, 24, 32, 48, plus one large step for the signature
Radiusone family chosen for the tone (sharp 2 to 6, soft 10 to 16, or round), not 8/12/16 by habit
Type1 to 3 families with the product's language (Vietnamese needs its subset); a display face with character for numbers and titles, a plain body face; a real scale contrast (the hero number 3 to 5 times the body); line height 1.4 text, 1.1 display; text-wrap: balance on titles
Elevationgroup by space and alignment first, a hairline second; a container only for a real separate object
Motiona few staged moments, not micro-animation everywhere; honour prefers-reduced-motion

5. Layout and hierarchy

  • Mobile first, 8 px grid, 16 to 20 px side gutters; group by proximity. Flex or grid with gap, never siblings spaced by margins.
  • One primary action per screen, in the lower half on a phone; tap targets at least 44 px; no text under 12 px. In every state (an alert, an error, a banner) that action stays on the first phone screen (844 px tall).
  • The title and the main action are what the eye finds first, by size, weight and colour, not borders around everything.
  • A web view differs from the phone on purpose (a second column, a side pane), not a stretched phone. Fluid roots (max-width, %, rem); a grid of cards is repeat(auto-fit, minmax(min(280px, 100%), 1fr)).
  • Diagrams: a flowchart of separate placed elements, not one big <svg>; <svg> is for an icon, a logo or a chart.

6. Built to be used

  • Real <button>, <a href> and <input> with a <label>, even in a static mockup; icon-only buttons get aria-label; decorative SVG is aria-hidden="true". Icons are inline SVG with one stroke width (1.75 to 2). Never emoji as icons.
  • No fake phone chrome: no painted status bar ("9:41", battery), no fake keyboard; the device draws them.
  • Draw the states, not only the happy screen: pressed, selected, disabled, empty (say what to do next), loading, error, offline, long text.
  • Text contrast at least 4.5:1 (3:1 from 24 px); grey captions and white text on a fill fail most often. Colours told apart differ in lightness, not only hue. Never colour alone for a state: add an icon or a word. Visible focus.

7. The canvas is a design document

A canvas that designs a product or a feature is read like a document, not browsed like a gallery of screens. (One frame asked for, or a change to one, stays a frame.)

  • Pages, in order: Directions (a new product), Wireframes (when asked), Foundations, Components, then one page per area of screens ("Screens · Today"), set with "page" in frame.json. Each page opens with a title label.
  • Wireframes show structure before style: every screen in greys, real labels, no colour or type choices, side by side in the order of the flow, with the steps between them; a caption names what each screen must answer. The main element (the signature, a 3D hero) is a grey line silhouette at its true size and place, never a hatched box with a label: its shape, what it overlaps, its states as dashed outlines, and its motion as arrows with timing ("↻ 20 s", "↑ 1,4 s"). Secondary pictures stay labelled boxes.
  • Every board is a spec board: a wide page (1440 to 1800 px, "fit": true) that opens with a small uppercase eyebrow (FOUNDATIONS, SCREEN · V1), a title and one or two sentences on what it is for and what it decides. Then lettered sections (a. Type scale, spacing, radii, c2. The six states of a reminder card), each with a heading saying what it shows and a one-line lead.
  • States side by side: the states of one thing in one row, each under a short label (Empty · first day, On track, Behind · after 3 pm, Goal met, Offline), drawn at the same size so only the state changes. Never one frame per state. A phone screen is a 390 px wide panel inside the board, real markup, not a picture.
  • A caption beside each decision: one or two muted lines saying the rule and why ("Visual size can be smaller than the target. The tappable area never overlaps its neighbour."). The reasoning lives on the canvas, not only in a comment.
  • Foundations first, at full depth, so a developer could build from that one board. Every value named by its use: the type scale as Card title / 15 / 600 / lh 1.35 under a sample in the real face; spacing as 12 card padding, 16 side gutter; radii as 12 card, button, input; touch targets (44 × 44 minimum); neutrals and brand colours with their job; status colours with their rule ("Dot colour only. The name always sits beside it."); then the base components with their states in a row (a button: primary, pressed, secondary, disabled, loading, destructive; an input: default, focus, error, disabled).
  • Components next: each repeated piece (a card, a row, a banner, a sheet) with its states and variants side by side.
  • Screens last, built only from those foundations and components: the screen in its main state, then a section of its states, then the web view at its own width.
  • Motion and illustration, when asked: inline SVG drawn for this product, and motion shown both running (a looping demo or a Play button in the board) and as key frames side by side with their timing and easing (craft.md).
  • A clickable prototype is extra: only when asked, as a last page of phone-size frames joined by arrows (frames share localStorage, so one can carry data to the next).

8. Avoid

The machine-made look (craft.md has each with what to do instead): every block a white rounded card on a tinted ground, an icon in a pale circle on every row, an alert as a tinted box with an icon, a bold title and a paragraph, a pill with an icon on every status, a row of equal buttons over four tab icons, a gradient wash, a left accent border, emoji. Also: lorem ipsum, invented numbers, low density dressed up as calm, text under 12 px, one-off pixel values, a frame per state, the same look as last time.

9. Changing an existing design

A small, targeted change (a word, a colour, one element) changes only that: every other layout, size, colour and word stays exactly as it is. If a broader change would help, finish what was asked and suggest the rest. A redesign or a new direction is different: then make the substantial changes. Follow the design's own vocabulary when you add to it.

10. Check before you send

  1. Squint: on each screen, are the title and the main action what you see first?
  2. Is every gap, size and colour a token, and is the direction (restrained or bold) the one you named?
  3. Do the boards read as a document: pages in order, a title and purpose on each, lettered sections, states side by side with labels, a caption by each decision, foundations a developer could build from?
  4. Does each screen hold at the width it is for (390 for a phone panel, 1280 and 820 for a web view)? No short label broken over two lines, nothing cut off, the main action on the first phone screen in every state.
  5. States there, copy real, nothing added that nobody asked for? Does it match what was asked, and nothing else changed?
  6. The thumbnail test: shrink the main screen to 120 px wide. Is the signature still there, and would it still look right under another app's name? Count the tells from craft.md; fix each, then look again.
  7. Beside the exemplars: is the presentation at their level of structure and finish?

11. Hand it over

New canvas: a folder with one sub-folder per frame (index.html, optional frame.json), then docs_write(canvas_create, owner_kind, owner_id, title, path); the owner (a ticket or a Docs page) must exist first. Put the returned markdown line on its own line in that ticket or page (manage_ticket or docs_write(update)).

  • Order and names: frames are placed left to right in folder-name order on their page, so name the folders 01-foundations, 02-components, 10-today; the number is not part of the shown name.
  • frame.json: {"name", "page": "Foundations", "size": "phone"|"tablet"|"desktop"|[w, h], "fit": true, "x", "y"}. Frames with the same page share a page tab, in the order they first appear. A long board is cut off by a fixed height: set "fit": true and the editor measures the page once, or give [w, h].
  • canvas.json at the top: labels ({"text", "page", "x", "y", "size"}, a page's title), notes ({"id", "text", "near": "<frame>", "page", "x", "y", "color", "w", "h"}; a note near a frame sits under it, on its page) and arrows ({"from", "to", "label"}, a frame folder name or a note id). A wrong reference is reported before anything is created.
  • A spec per screen on Docs, for whoever builds it: one page per screen (or a section each on one page) with what it is for, a screenshot of each state (![Today · behind](shots/today-behind.png), a file in your working folder), the tokens and components it uses, sizes and behaviour, the copy of every state, open questions, and the board it is on.
  • Change to an existing frame: write the changed files to a folder and docs_write(canvas_update) with a one-sentence note on the design decision; the board changes at once.
  • Leave canvas_comments for open questions, each pinned with at (a CSS selector, or text=Add task). Give your assumptions in one line.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.