Ships the headline capability: point tokenscout at a URL and get design tokens. Stacked on #9 (base will retarget to main once #9 merges).

What

• Workspace split so the core stays zero-dependency: published core moves to packages/core (name stays tokenscout, no consumer breakage); root is now a private npm workspace with a shared tsconfig.base.json.
• New @tokenscout/extract package, Playwright as a peer dependency:
  • extractSite(url, { breakpoints, top }) drives headless Chromium, reads computed styles (color / background / border, font-size, margin/padding/gap) at each breakpoint with optional same-origin crawling, returns PageExtract[].
  • extractTokens(url, opts) chains that into the core's assembleTokens for a one-call URL → DTCG flow.
  • Aggregation is a pure, unit-tested harvest(); the browser-side collector stays thin. A real-Chromium smoke test runs against a local fixture and asserts the end-to-end path.
• CI: split into a fast core job and a Chromium-backed extract job. Release: publishes the core from packages/core with provenance.
• Docs updated to match reality (no more "planned" for things that now exist): README live-URL usage + install + roadmap + limitations, ARCHITECTURE real two-package layout + CI, ROADMAP Phase 3 in progress, CHANGELOG.

Verification

• npm run build, typecheck, lint, format:check: clean.
• Tests: core 57/57, extract 3/3 (incl. the live-browser smoke test). Zero em-dashes repo-wide.

Usage

import { extractTokens } from "@tokenscout/extract";
const tokens = await extractTokens("https://example.com", { breakpoints: [1280, 375], top: 1 });

Not in this PR (still roadmap)

Image/asset harvesting and animation capture. Sitemap-driven discovery and tech-stack profiling.

🤖 Generated with Claude Code