agentleFS
BarganConstantinBarganConstantin/agent-skills/skills/figma-design-system/SKILL.md

figma-design-system

Skill4 starsChanged 15 days agoMITView on GitHub

Designs a product's UI in Figma before any code, the way a senior product designer would. It works through the Figma MCP in one file. Research first, then foundations (variables for Light and Dark, type, icons), a component library with every state, measured page anatomy, every screen and state in both themes, motion, an independent design critique, fixes, a visual check of every frame, and a written record with the open questions for the owner. Use when the user says "design it in Figma first", "build a design system in Figma", "every screen in light and dark", "do the UI before we code", "design review in Figma", "fă design system în Figma", "construim UI-ul în Figma", or gives a Figma file and asks for a product's screens.

2 other projects use a version of this.

Adopted by

Most recently taken up by

  • Other versions2 projects use a version of this.

    Word for word · 0

    None found.

    Edited · 0

    None found.

    Wrote their own version · 2

    1. BreRoz/ClaudeSkills★ 0same name
    2. flntfnd/claude-skills★ 0same name

    How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.

  • In this repository4 other files it gives its agents.

The file

Show its rules and text

We split this skill into the rules it gives an agent

19 rules. About 0 are common, 18 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈18

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    every screen and state in Light and Dark.

    line 14
  2. Distinctive:

    Design first, code later: Write no product code until the owner approves the design.

    line 22
  3. Distinctive:

    One Figma file: Use only the file the user gives.

    line 23
  4. Distinctive:

    Use the brand kit exactly as delivered: Never recolour or re-trace a logo, mascot or icon.

    line 28
  5. Distinctive:

    When the kit has no token for a need, derive a step from its own ramps and mark it *derived*.

    line 29
  6. Distinctive:

    All colour is a variable: No screen holds a hard-coded colour.

    line 31
  7. Distinctive:

    Check by looking: After every change, take a screenshot of each frame it touched and look at it.

    line 33
  8. Very specific:

    Before the first use_figma call, load the Figma MCP's figma-use skill.

    line 41
  9. Distinctive:

    Read references/plugin-api.md.

    line 43
  10. Distinctive:

    Read the project's product notes, decisions and brand kit.

    line 45
  11. Distinctive:

    Use one sample dataset across all screens, so the numbers on an overview, a list and a detail page agree.

    line 112
  12. Distinctive:

    Put an annotation under any frame that needs a reason, a proposal or a pending decision.

    line 116
  13. Distinctive:

    Never animate an action started from the keyboard.

    line 127
  14. Distinctive:

    Never move rows while someone reads them: new items wait behind a pill.

    line 128
  15. Distinctive:

    Start a reviewer with a fresh context: another agent, or a person.

    line 135
  16. Distinctive:

    Use the templates in references/records.md:

    line 167
  17. Distinctive:

    When the user asked for notifications, also send one at each milestone and whenever you need them.

    line 180
  18. Distinctive:

    Before you build something you think the product does not need, say so, even when it was asked for.

    line 181
  19. Distinctive:

    Write all copy in plain, short sentences: one idea per sentence, active voice, no filler.

    line 183

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “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

figma-design-system: design the whole UI in Figma first

The output is a Figma file that a developer can build from without guessing, and a short written record in the project's repository. The file holds:

  • the foundations;
  • the component library;
  • every screen and state in Light and Dark;
  • the motion.

Work in phases. Finish each phase's exit check before you start the next one. Do not ask the user to confirm between phases. Ask only for decisions that are theirs.

Hard rules

  1. Design first, code later. Write no product code until the owner approves the design.
  2. One Figma file. Use only the file the user gives. Never publish it to the Community or share it by a public link. Every other deliverable is a local file in the repository.
  3. Never invent a product decision. Draw a value that no project document records as an annotation of kind Proposal or Pending decision, and put the question to the owner. Screen copy keeps to the project's documented facts: status codes, limits, durations, retention, roles.
  4. Use the brand kit exactly as delivered. Never recolour or re-trace a logo, mascot or icon. When the kit has no token for a need, derive a step from its own ramps and mark it derived. Then ask the designer to confirm or replace it.
  5. All colour is a variable. No screen holds a hard-coded colour. Light and Dark come from the same frames through variable modes.
  6. Check by looking. After every change, take a screenshot of each frame it touched and look at it. A script that succeeded proves nothing about how the frame looks.
  7. Annotations are review notes, not product UI. They sit under the frames as a component with three kinds: Note, Proposal, Pending decision.

Session setup

  1. Get the file key from the URL. Call create_session once and pass its ID on every call.
  2. Before the first use_figma call, load the Figma MCP's figma-use skill. Load figma-generate-library too when you build the library.
  3. Read references/plugin-api.md. It has the script preamble, the recipes, and the traps that cost hours.
  4. Read the project's product notes, decisions and brand kit. List the facts the copy must keep to.

Phases

1. Research

Study five to eight products that solve the same job. Look at their layout, density, states, copy and motion. Use a screen archive MCP when you have one. Write a brief in the repository, in reports/<date>-<topic>-design-research.md, with:

  • principles, each with the evidence behind it;
  • the states that each screen family needs (start from references/states.md);
  • the questions for the owner, numbered (Q-O1…);
  • the questions for the brand designer (Q-D1…).

Exit: every later design choice can point to a line in the brief.

2. Foundations

Make these collections, in this order:

  1. Primitives: the kit's colours, with derived steps marked.
  2. Semantic UI tokens with Light and Dark modes: bg, border, text, action, status, chart, overlay.
  3. Dimensions: spacing and radius.
  4. Motion: durations and easings.

Then add text styles (UI and mono, tabular numbers), effect styles (elevation and focus ring), and one icon set in one line style.

Check the work:

  • contrast for text and controls, in both modes;
  • chart palettes with a colour-vision validator, in both modes;
  • that status hues never double as chart series.

Exit: the Foundations page documents every token, and no value outside the variables is needed.

3. Components

Build each component with all its states as variants. That means default, hover, pressed, focus, disabled, loading, error and selected, as they apply. Bind everything to tokens, so each component works in both modes. Group the library: status and identity, actions, inputs, navigation, feedback and overlays, data.

Exit: every repeated element on a screen is an instance. A one-off frame on a screen is a defect: make it a component or reuse one.

4. Page anatomy

On a Patterns page, measure the shell for developers:

  • gutters, sidebar and bar heights, content padding;
  • table row heights;
  • overlay widths and the stacking order;
  • focus rings and target sizes;
  • breakpoints, as a proposal.

Exit: a developer can lay out a page from this page alone.

5. Screens

  • Give each flow one row on the Screens page, numbered 01, 02 and so on. Number its states 01b, 01c, then 03c2 for a state inserted later. Light sits on the left of each pair and Dark on the right.
  • Draw every state the brief lists: empty, loading, error, filtered with no results, success, confirmations, permission denied, not found, first run, and the phone layout.
  • Use one sample dataset across all screens, so the numbers on an overview, a list and a detail page agree.
  • Build and check the Light frame first. Then make the Dark copy by setting the frame's variable mode (see the recipe).
  • Put an annotation under any frame that needs a reason, a proposal or a pending decision.

Exit: every flow is clickable as a prototype, and every state from the brief has a frame in both themes.

6. Motion

  • Storyboard the few signature moments. Put every other transition in one table: trigger, property, duration, easing.
  • Entering elements use a strong ease-out. Elements that move on screen use ease-in-out. Never use ease-in.
  • Never animate an action started from the keyboard.
  • Never move rows while someone reads them: new items wait behind a pill.
  • Give each moment a reduced-motion version.

Exit: every animation has a written purpose.

7. Independent critique

Start a reviewer with a fresh context: another agent, or a person. Give it the file, the list of frames and the project's facts, and never your own conclusions. Use the brief in references/critique.md.

Sort the findings into batches:

  • A, copy and facts;
  • B, sample data and tables;
  • C, structure and new states.

Fix one batch at a time. Record what the reviewer said to keep, and keep it.

Exit: every finding is fixed, or written down as open with a reason.

8. Visual check

For each frame you touched, take a screenshot in Light and in Dark. Tile the screenshots when there are many, then read them as a designer would. Look for:

  • clipped or wrapped labels;
  • misaligned rows across cards;
  • one-off styles;
  • leaks between environments;
  • stale numbers;
  • overlapping frames on the canvas.

Fix what you find, then take the screenshots again.

Exit: a full pass finds nothing new.

9. Record and hand-off

Use the templates in references/records.md:

  • A summary report in the repository: counts, the screens table, what the design does not decide, how it was checked, and what happens after approval.
  • A review issue for the owner: how to review in about 20 minutes, the questions as checkboxes, and the proposals to accept or reject.
  • An issue for the brand designer: the missing tokens and the derived steps.
  • On approval: export the approved frames as PNG to the repository's design/ folder. Then build from the tokens and the components.

Working style

  • Long work continues without a pause between phases. Send a short status line from time to time. When the user asked for notifications, also send one at each milestone and whenever you need them.
  • Before you build something you think the product does not need, say so, even when it was asked for.
  • Write all copy in plain, short sentences: one idea per sentence, active voice, no filler.

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.