agentleFS
Sign inStart free
udecodeudecode/better-convex/.agents/skills/design/SKILL.md

design

Skill452 starsChanged 23 days agoApache-2.0View on GitHub

Design and implement kitcn UI using live www/example tokens and primitives, accessible states, honest data formatting, and browser proof.

We split this skill into the rules it gives an agent

18 rules. 0 are common, 16 are distinctive, and 2 are very specific to this project.

  • 0

    Common

    Most of the 174 edited versions we read keep these.

  • 16

    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:

    Treat live tokens and components as the palette.

    line 24
  2. Distinctive:

    Use semantic color, spacing, radius, type, and motion tokens.

    line 29
  3. Distinctive:

    Prefer existing shadcn-derived primitives and composition patterns.

    line 32
  4. Distinctive:

    Keep docs and example variants consistent when they express the same product concept, while preserving their distinct runtime ownership.

    line 34
  5. Distinctive:

    Start with the user's decision:

    line 39
  6. Distinctive:

    Do not render zero as missing, missing as zero, or inferred data as confirmed.

    line 61
  7. Distinctive:

    Use semantic elements before ARIA.

    line 66
  8. Distinctive:

    Keep keyboard order aligned with visual order.

    line 68
  9. Distinctive:

    Do not encode status only by color.

    line 71
  10. Distinctive:

    Use product language already established in docs and examples.

    line 76
  11. Distinctive:

    Prefer direct verbs and concrete nouns.

    line 77
  12. Distinctive:

    Never delay the user's next action for decoration.

    line 86
  13. Distinctive:

    Write the state and interaction matrix.

    line 92
  14. Distinctive:

    Test responsive, keyboard, reduced-motion, loading, empty, error, and permission states.

    line 95
  15. Very specific:

    Use Browser for route/DOM/responsive/screenshot proof.

    line 97
  16. Very specific:

    Run focused tests and React review rules, then autoreview.

    line 99
  17. Distinctive:

    All states are honest and actionable.

    line 104
  18. Distinctive:

    No fake data or silent permission assumption is needed for the screenshot.

    line 107

“Common” is measured: kept by at least 6 in 10 of the 174 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

Kitcn Design

Design from the live subject, data, and interaction. A polished shell around fake states is not done.

Read First

  1. VISION.md and the named product/API source.
  2. The live route and its data contract.
  3. www/app/global.css, www/components/ui/**, and nearby documentation UI.
  4. example/src/app/globals.css, example/src/components/ui/**, and the matching example feature.
  5. Existing loading, empty, error, permission, and mutation states.

Treat live tokens and components as the palette. Do not introduce an isolated design language in a feature file.

Tokens And Primitives

  • Use semantic color, spacing, radius, type, and motion tokens.
  • Extend the token owner when a reusable need is real; do not hide raw values behind one-off aliases.
  • Prefer existing shadcn-derived primitives and composition patterns.
  • Change primitive behavior only when every consumer benefits.
  • Keep docs and example variants consistent when they express the same product concept, while preserving their distinct runtime ownership.

Subject And Hierarchy

Start with the user's decision:

  • What is the primary object or action?
  • Which facts change the decision?
  • What is supporting context?
  • What is optional detail?

One view gets one dominant hierarchy. Remove competing cards, repeated labels, and decorative metadata that do not improve comprehension.

Honest States

Design every state before declaring the happy path complete:

  • initial/loading and background refresh;
  • empty versus filtered-empty;
  • missing setup or permission;
  • inline validation and server rejection;
  • mutation pending, success, partial failure, and retry;
  • stale or unavailable external data;
  • destructive confirmation and irreversible completion.

Do not render zero as missing, missing as zero, or inferred data as confirmed. Label samples and previews. Preserve the user's input after failure.

Accessibility

  • Use semantic elements before ARIA.
  • Give controls visible labels or accessible names.
  • Keep keyboard order aligned with visual order.
  • Provide focus visibility, escape behavior, and focus restoration for overlays.
  • Announce important async success/error changes without noisy repetition.
  • Do not encode status only by color.
  • Respect reduced motion and usable target sizes.

Copy And Formatting

  • Use product language already established in docs and examples.
  • Prefer direct verbs and concrete nouns.
  • Explain consequences before destructive or expensive actions.
  • Format dates, numbers, identifiers, and durations from typed values with the intended locale/timezone; do not pre-format them in the data layer.
  • Empty copy should say what is absent and the useful next action.

Motion

Motion explains state change. Keep it short, interruptible, and restrained. Never delay the user's next action for decoration. Skeletons should preserve layout; spinners belong to bounded operations.

Process

  1. Inspect the running surface and representative data.
  2. Write the state and interaction matrix.
  3. Reuse or adjust live tokens/primitives.
  4. Implement the smallest coherent component ownership.
  5. Test responsive, keyboard, reduced-motion, loading, empty, error, and permission states.
  6. Use Browser for route/DOM/responsive/screenshot proof. Use Chrome or Computer only when native browser/OS behavior requires it.
  7. Run focused tests and React review rules, then autoreview.

Definition Of Done

  • The real subject and primary action are obvious.
  • All states are honest and actionable.
  • Tokens and primitives have clear owners.
  • Keyboard, focus, names, contrast, and motion are proven.
  • No fake data or silent permission assumption is needed for the screenshot.
  • Browser evidence covers the route and relevant breakpoints.
  • Docs/example source ownership remains correct.

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.

Check your setup, free