chore: never hard-wrap markdown, update @tenphi/tasty to 3.3.1

[tenphi]

Markdown is never hard-wrapped

One paragraph is one line, in every .md and .mdx file — docs, AGENTS.md, specs, changesets. Hard wrapping is what made these files awkward to edit and to review: a one-word change reflows the paragraph, so the diff reads as a full rewrite and every editor disagrees about where the breaks belong.

Enforced rather than asked for:

• proseWrap: 'never' in .prettierrc.cjs, with embeddedLanguageFormatting: 'off' for *.md/*.mdx — code fences in docs are illustrative snippets, not source files, and formatting them would rewrite deliberate pseudo-code.
• *.{md,mdx} added to the prettier / prettier:fix scripts (so pnpm lint and CI check it) and to .lintstagedrc (so a commit fixes it).
• New .prettierignore keeps generated output out: CHANGELOG.md (changesets owns it), dist, storybook builds, the lockfile.
• The rule itself is written down in docs/rules/coding.md, pointed at from AGENTS.md, CONTRIBUTING.md and the add-changeset command.

97 Markdown files are unwrapped in this pass. pnpm build-docs builds clean, so every reformatted .docs.mdx still compiles.

One file needed a manual touch: a bullet in Board.docs.mdx mixing inline JSX with wrapped prose made Prettier oscillate between two indentations — check and write disagreeing forever. Joining it into one line settles it, and the condition needs already-wrapped source to reproduce.

A latent bug the unwrap surfaced

no-redundant-default-prop's registry is parsed line by line out of *.docs.mdx, and a prop only becomes a candidate when its bullet carries the (default: …) annotation on the same line as the prop name. A hard-wrapped bullet hid its own default. Three come back — ItemTable's size and summary, and Pagination's summary — each proven against the component it belongs to by the usual generator, not just parsed. defaults.generated.ts is regenerated; the note in docs/rules/eslint-plugin.md now says why wrapping breaks it.

Tasty 3.3.1

Straight from 3.1.0 to 3.3.1, and two of the three changes reach this repo.

Color tokens no longer emit a decomposed companion variable, and colorSpace is deprecated. A #name token used to declare --name-color plus a companion — --name-color-rgb here, since Root configured colorSpace: 'rgb' — and the companion is what's gone; --name-color is emitted exactly as authored. Opacity has used relative color syntax since 3.1.0, so nothing inside Tasty still needed the channels. Two things had to change here: Root stops setting colorSpace (it now warns in dev and does nothing), and the two rgb(var(--primary-color-rgb) / .2) selection rules in GlobalStyles become oklch(from var(--primary-color) l c h / .2). Left alone they would have silently stopped resolving — the PrismCode selection highlight would just have gone transparent. The replacement is the same pattern Tasty itself emits (--scrollbar-thumb-color), so it adds no browser floor. This is why the changeset is minor: a consumer reading a companion in their own CSS is broken the same way. Computed colors also serialize in the space the token was authored in now, so getComputedStyle(el).backgroundColor returns oklch(…) where it returned rgb(…) — same color, but any test parsing those strings by hand has to stop assuming sRGB syntax. That is what broke the DataTable browser tests below.

The inherited color is readable as a color through $current-color. #current stays the currentcolor keyword, so it still resolves against the element reading it — 3.3.0 briefly made it var(--current-color), which broke the current.primary disabled chip (the fill stopped inheriting the fade its own color applies and rendered at full strength); 3.3.1 restores the keyword and publishes the variable alongside instead, for the cases the keyword cannot serve. Verified on 3.3.1: the disabled chip measures oklab(0.571 0.0173617 -0.0955352 / 0.4), byte for byte its 3.1.0 value. Nothing in the kit had to change.

Opt-in batched injection (from 3.2.0): configure({ batchInjection: true }) + <TastyBatchProvider> collapse a commit's stylesheet writes into one flush, for trees that measure layout during render. Off unless an app turns it on.

Size, both sides rebuilt locally against 3.1.0: All 509.94 → 508.53 kB (−1.41 kB), Button-only 124.03 → 121.76 kB (−2.27 kB). 3.2.0's batching cost 0.68 kB on the Button entry; 3.3.0 then gave back 2.96 kB by deleting the color-space decomposition. Both budgets are unchanged at 515 kB and 125 kB, with the readings and the reasoning recorded in .size-limit.cjs as that file asks.

Verification

• pnpm test — 2093 passed, 1 skipped, 0 failed
• pnpm test:browser — 123 passed (the DataTable color helpers now resolve any color syntax through a canvas instead of parsing rgb(...) by hand)
• pnpm lint — clean
• pnpm build-docs — clean
• pnpm size — 508.53 / 515 kB and 121.76 / 125 kB
• Rebased on main (3e5e3fee); the Tooltip docs it brought in are unwrapped too

🤖 Generated with Claude Code

---

Note

Medium Risk
The Tasty minor upgrade changes emitted CSS variables and computed color serialization, which can break consumer styles and tests that assumed -rgb companions or rgb(...) strings; the in-repo migration is small but the dependency bump is library-wide.

Overview
Markdown prose is enforced as single-line paragraphs across the repo: Prettier gets proseWrap: 'never', MD/MDX overrides, .prettierignore, lint-staged + pnpm lint coverage, and documented rules in CONTRIBUTING / docs/rules/coding.md. ~97 .md/.mdx files are reformatted to match.

@tenphi/tasty moves from 3.1.0 to 3.3.1 (minor changeset). The kit drops configure({ colorSpace: 'rgb' }) in Root and rewrites Prism selection highlights in GlobalStyles from rgb(var(--primary-color-rgb) / …) to relative oklch(from var(--primary-color) …) so they keep working now that tokens no longer emit -rgb companions. Consumers who still reference companion variables in custom CSS need the same migration; computed color strings may serialize as oklch/oklab instead of rgb.

Docs unwrap fixes the ESLint defaults registry: line-based parsing of (default: …) in *.docs.mdx regains ItemTable size/summary and Pagination summary; defaults.generated.ts is regenerated and docs/rules/eslint-plugin.md explains the coupling.

Tooling-only follow-ups: size-limit comment updates for the Tasty bundle delta, changeset/agent guidelines to write unwrapped summaries, and widespread .docs.mdx whitespace/formatting with no API changes beyond the Tasty/CSS items above.

^{Reviewed by Cursor Bugbot for commit 1035d18. Bugbot is set up for automated code reviews on this repo. Configure here.}

[changeset-bot]

🦋 Changeset detected

Latest commit: 1035d18

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 1 package

Name	Type
@cube-dev/ui-kit	Minor

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
cube-ui-kit	Ready	Preview	Aug 24, 2026 4:25pm

[github-actions]

📦 NPM canary release

Deployed canary version 0.0.0-canary-20121c0.

[github-actions]

🧪 Storybook is successfully deployed!

• 👀 Review changes: https://www.chromatic.com/build?appId=626ee6db91d1c8004a912249&number=3961
• 👨‍🎨 Preview storybook: https://626ee6db91d1c8004a912249-zubxsnhbzt.chromatic.com/

[github-actions]

🏋️ Size limit report

Name	Size	Passed?
All	497.02 KB (-0.29% 🔽👏)	Yes 🎉
Tree shaking (just a Button)	118.91 KB (-1.84% 🔽👏)	Yes 🎉

Compared against main at 3755065 — run 32750166384, 2026-08-24T16:19:32Z.

To see which modules changed, download the size-limit-statoscope-report artifact from this run and open report.html.