Index of every page on this site: llms.txt
All skills
vercel-labs avatar

/ui-sdk-write-code

@fb6945a
by Vercel Labsvercel-labs/ui6.3k stars
218
Run it yourself

Writes and edits ui-sdk code (core, the React, Vue, Svelte, Solid and web component adapters, examples, specs and scripts) the way an expert in that framework would, then checks it with the linters, the framework expert's checklist and a cold read. Use whenever you write or change code in this repo, before you call the change done.

  • 1 file
  • 5.9 KB
  • Updated last week
  • GitHub
Use this Skill: https://skilld.dev/gh/vercel-labs/ui/ui-sdk-write-code

Nothing lands on disk. Nothing to clean up.

Fork this Skill

Edit a local copy. It keeps the author and licence.

SKILL.md

≈88 tokens for metadata: the name and description. ≈1.4k when used: this file.

Description uses 4.4% of example budget

Before choosing a Skill, your Agent reads its name and description. All available Skills share that space.

  • A shorter description leaves more room for other Skills. This entry exceeds our 1% size suggestion.

In our Claude Code example, all Skill names and descriptions share 8,000 characters. This Skill uses ≈353 characters, or 4.4%.

The 1% threshold is a size suggestion. Longer descriptions can still fit.

Your model, settings, and other Skills decide how much text your Agent can read.

Example settings and source

The example uses a 200k-token context and default Claude Code settings. The count includes the name, description, separators, and when_to_use when present. Codex also counts local file paths.

Skit's source and limits: Codex 0.160.1, Claude Code 2.1.292.

Write ui-sdk code

The bar: an expert in the framework reads the file and says "this is how I'd write it", and anyone else reads it top to bottom without stopping. Readable first, then fast, never clever.

1. Find the twin

Every piece of code here has a twin that already does it well. Open it before you write a line, and copy its shape, names and order.

You're writing Start from
A core component packages/core/src/components/tabs/tabs.ts, or menu/ once it passes about 400 lines (menu.ts, menu-api.ts, menu-parts.ts)
A popup packages/core/src/components/popover/popover.ts and its adapters, on createPopup in foundation/popups/
A React, Vue or Solid part The same component's file in packages/<framework>/src/, on createParts
A Svelte part packages/svelte/src/tabs/
A web component packages/web/src/tabs.ts, on createParts
An example The same example in another framework: examples/components/<name>/<framework>/
A spec tests/specs/tabs*.spec.ts

If two twins disagree, the shape is wrong in one of them. Say so instead of picking one.

Copy a twin's shape, not its comments. Older files carry more comments than the rules below allow.

2. Load the framework

Read the expert's knowledge file for every framework you touch: .agents/experts/<framework>/knowledge.md, part 1 and "Common mistakes to flag", and its decisions.md. It's pinned to the versions this repo installs, so trust it over memory. Write in the current idiom of that version:

  • React 19: pure render, no forwardRef, ref as a prop, no effect where an event handler or a derived value belongs, code the React Compiler compiles without a bailout.
  • Vue 3.5: <script setup lang="ts">, v-model on parts, defineModel when a component you write takes one, typed defineProps and defineSlots, no options API.
  • Svelte 5: runes only ($state, $derived, $props, $bindable), snippets, onclick attributes, attachments. No $:, on:, <slot> or createEventDispatcher. Props typed with an interface Props. Outside packages/svelte, declaration tags ({const x = $derived(...)}) instead of {@const}.
  • Solid 2: props read where they're used, never destructured; derive with memos, not effects; Solid 2 APIs only.
  • Web components: platform first. Options reflect as attributes the way built-ins do, public events are composed and don't bubble (listen with { capture: true }), and the user's markup stays in light DOM.

3. Write it

The rules in contributing/code.md hold everywhere. The ones that make code look generated when missed:

  • One shape. <Name> root, <Name><Part> per slot, render to change the element, never as. Same names and the same props in every framework.
  • Names in full words that say what a thing is. Booleans read as questions.
  • Exports listed once at the end. Inline types. No return type annotations.
  • No comment by default. Write one only when a reader would get the code wrong without it: a browser bug, an APG rule, a trade-off. One to three lines. Never a comment per declaration, and never one that explains the design or the data flow. Examples have none.
  • No defensive code for cases that can't happen. Types already rule them out.
  • No helper for one caller, and no wrapper that only renames.
  • Behavior lives in core. An adapter that grows a listener, a focus call or an aria-* attribute is in the wrong layer.

4. Check it

Run each and check the exit code, not the output:

pnpm lint
pnpm format
pnpm typecheck
pnpm --filter <package> test
pnpm --filter examples build   # when you touched examples
pnpm test:pr                   # the specs your change reaches

Then run each framework's own lint rules on the files you touched: npx eslint <files> (pnpm lint:idioms runs them on the whole repo, as CI does). Fix what each one reports and run it again until all pass. Older errors are listed in eslint-suppressions.json; when you fix one, run pnpm lint:idioms --prune-suppressions.

5. Read it cold

Read the whole diff as the framework's expert would, using the "Review checklist" in each knowledge.md you loaded, and as a teammate who has never seen it:

  • Would an expert in this framework write it this way? If you can name a more idiomatic version, write that one.
  • Is every name clear without its comment?
  • Is anything duplicated that exists elsewhere?
  • Does any function do two things, or any file hold two ideas?
  • Any allocation on a hot path, or a layout read after a write?

6. Unslop

Run ui-sdk-unslop on your diff. The change isn't done until it has. Before merging a change to an adapter, core, the specs, examples or framework docs, run ui-sdk-expert-review.

Source: SKILL.md on GitHub

skilld matched fixed text patterns in SKILL.md and file names. Patterns miss obfuscated code.

skilld run checks every file with the same patterns. It asks for approval before it loads a Skill with a behavior marked Needs approval.

No third-party reports yet.

Signed by skilld at fb6945a. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub 14 hours ago.

Activeupdated last week
metadata
{
  "internal": true
}

README badge

README badge for vercel-labs/ui/ui-sdk-write-code