feat!: prepare 0.7.0 release with peer-dep migration and consumer polish

[zhawtof]

Summary

Exits the 0.6.x-alpha line and tightens the consumer-facing surface ahead
of the first stable 0.x release. The big change is the dependency model:
a narrow set of runtime deps move to peer so shadcn/Radix and lucide
consumers dedupe instead of getting duplicate copies.

Why

block-kitchen is built to be embedded into someone else's app, and a
quick audit of how it handles that today surfaced a few rough edges:

1. Duplicate Radix / lucide trees in consumers' node_modules
   (Radix in particular risks TooltipProvider/PopoverContext mismatch
   when two copies are present).
2. Subpath imports missing — a backend that only needs toSlackBlocks
   was forced through the root entry and its full React component tree.
3. No quick palette filtering — the only way to remove an image or
   table block from the default palette was to rebuild the whole array.
4. Tailwind cascade collisions — package utilities and consumer
   utilities both compiled into @layer utilities, leaving conflict
   resolution to source-import order.
5. No SSR / framework guidance — README didn't mention 'use client',
   portal placement, or where to import the stylesheet.
6. Alpha versioning with no documented stability signal.

This PR addresses all six.

Breaking change

A narrow set of dependencies moved from dependencies to
peerDependencies — just the ones a typical consumer is likely to
already have, where dedupe gives them a real win:

• react, react-dom
• @radix-ui/react-{dialog,label,popover,radio-group,slot,tooltip} —
  shadcn/ui audience already has these, and peer avoids the React-
  context split bug from two Radix copies.
• lucide-react — universal in modern React UI stacks.

Everything else stays as a regular dependency. @dnd-kit/*,
@tiptap/*, slack-blocks-to-jsx, and
@tightknitai/slack-block-kit-validator are specialty — most consumers
don't already have them, so peer would only add version-skew surface
with no dedupe upside.

Modern package managers (pnpm 8+, npm 7+, yarn berry) auto-install
peers — most consumers won't need to change anything. Older tooling
can use the explicit command added to the README:

pnpm add @tightknitai/block-kitchen \
  react react-dom \
  @radix-ui/react-dialog @radix-ui/react-label \
  @radix-ui/react-popover @radix-ui/react-radio-group \
  @radix-ui/react-slot @radix-ui/react-tooltip \
  lucide-react

slack-web-api-client is now a devDependency (type-only consumer; stays
in tsup's external list).

What's in the PR

• feat(exports) — subpath entries:
  • @tightknitai/block-kitchen/helpers (4.5 KB) for headless backend
    usage of toSlackBlocks, encodeBlocksToString, decodeBlocksFromString.
  • @tightknitai/block-kitchen/palette for tools that need
    defaultPalette / legacyInputVariants / extraAlertVariant /
    buildVariantById without the builder.
  • Root entry still re-exports everything — no breaking change for
    existing imports.
• feat(palette) — new disabledBlockTypes?: SupportedBlockType[]
  prop on <BlockKitchen>. Filters at the variant level (sections that
  mix types keep variants whose types aren't disabled; sections that
  empty out drop entirely).
• refactor(styles) — package's compiled Tailwind utilities now
  live in named cascade layers (bk-theme, bk-utilities) so consumer
  utilities win on collisions without source-order luck. Layer order
  documented in the README for consumers who want explicit control.
• docs(readme) — bundlephobia badge; framework recipes for Next
  (App + Pages Router), Remix, Vite, Astro; portal-stylesheet pitfall
  callout; subpath-export examples; cascade-layer ordering snippet;
  disabledBlockTypes row in the props table; explanation of the
  narrow peer-dep set and the deliberate cut-off.
• chore(release) — package.json and .release-please-manifest.json
  bumped from 0.6.2-alpha.0 to 0.7.0. Pre-1.0 minor bump per
  release-please config (bump-minor-pre-major: true) signals the
  breaking peer-dep migration.

Test plan

• pnpm typecheck
• pnpm test — 217/217 passing
• pnpm build — clean, ESM + CJS + types for all three entries
• pnpm --dir demo install && pnpm --dir demo build — demo links to
  source and still builds against the new dep layout
• Demo dev server (pnpm demo:dev) renders palette + surface +
  templates panel with no console errors after the cascade-layer
  rename
• Reviewer to spot-check pnpm storybook
• Reviewer to spot-check dark mode (Mode: dark toggle in demo
  header)

Notes for reviewer

• The 0.7.0 bump in .release-please-manifest.json is intentional —
  it's the mechanism that gets the package out of the -alpha.X suffix
  on the next release-please run.
• Bundle isn't smaller (already-externalized deps stay externalized) but
  consumers' node_modules size drops meaningfully when they share
  Radix / lucide with the rest of their app. The biggest remaining
  bundle win — React.lazy per-block-editor code splitting — is
  deliberately deferred; it's a real refactor, not a release-prep fix.
• The bundlephobia badge will start populating once 0.7.0 lands on
  npm. Until then it'll render as "no data" on bundlephobia.
• Peer-dep set was narrowed mid-review from the initial broader cut
  (which also included @dnd-kit/*, @tiptap/*, slack-blocks-to-jsx)
  after deciding those don't carry the dedupe / context-mismatch
  justification that Radix and lucide do — they're specialty deps the
  typical consumer wouldn't already have, so peer would just add skew
  warnings.

🤖 Generated with Claude Code

[cloudflare-workers-and-pages]

Deploying with    Cloudflare Workers

The latest updates on your project. Learn more about integrating Git with Workers.

Status	Name	Latest Commit	Preview URL	Updated (UTC)
✅ Deployment successful! View logs	block-kitchen	cb6d2a2	Commit Preview URL Branch Preview URL	May 19 2026, 02:54 PM