- .github/workflows/bench.yml: manual (choose a set) plus weekly, installs the
browser, runs held-out and training sets, writes both tables to the run
summary and uploads bench/results as an artifact. Nothing is auto-committed.
- the report names the set it ran (a held-out run said bench/sites.json) and
says when it is the held-out set
- median time rounds to one decimal (it printed 30.349999999999998s)
- release.yml publishes with provenance when a vX.Y.Z tag matches package.json
- MCP serverInfo reports the package version instead of a hard-coded 7.0.0
- add src/mcp/index.js, which the ./mcp export pointed at but never existed
- test a stock SDK client handshake over stdio
Runs the full suite on Node 20 and 22 for every push to main and every
PR, installs Chromium for the extraction-failure test, and verifies the
plugin manifest version stays in sync. Gives auto-merge something real to
gate on.
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Adds .github/dependabot.yml (weekly npm + github-actions; minor/patch
grouped) and a dependabot-auto-merge workflow that enables GitHub
auto-merge (gh pr merge --auto --squash) for patch & minor bumps, leaving
major versions for manual review. Uses fetch-metadata; PR URL passed via
a quoted env var (no injection); does not check out PR code.
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
* docs: v7.0 spec and implementation plan
* chore: add @modelcontextprotocol/sdk for v7 MCP server
* feat(formatters): add DTCG v1 tokens with primitive + semantic + composites
* feat: use DTCG token format by default; legacy via --tokens-legacy
* feat(extractors): add stack-fingerprint for framework + Tailwind detection
Detects framework (next/nuxt/gatsby/remix/react/vue/astro/sveltekit/
shopify/wordpress/webflow/framer), CSS layer (tailwind with top-100
utility frequencies), and analytics providers from crawled page
signals (scripts, meta tags, class-name sample, window globals).
Crawler now emits rawData.light.stack carrying those signals.
design.stack lands on the extraction output.
* feat(extractors): CSS health audit (specificity, !important, unused, keyframes)
Uses Playwright CSS coverage to compute used/unused bytes, counts
!important usage, duplicate prop:value pairs, vendor-prefix hits,
selector specificity distribution, and catalogs @keyframes.
Crawler starts coverage before navigation and stops after the page
settles; gracefully degrades to empty coverage on unsupported targets.
scoring.js gains an additive cssHealth dimension (does not change
existing weighted overall score for back-compat).
* feat(a11y): remediation suggestions — nearest palette color passing WCAG
For every failing contrast pair emitted by the accessibility extractor,
search design.colors.all for a palette entry that clears the matching
WCAG threshold (AA-normal=4.5, AA-large=3, AAA variants).
Exposes design.accessibility.failingPairs (derived) and
design.accessibility.remediation (array of { fg, bg, ratio, rule,
suggestion }). Suggestion is null when no palette color qualifies.
* feat(extractors): semantic-regions classifier (nav/hero/pricing/footer/…)
Classifies page sections using landmark tag/role plus keyword and
layout heuristics into: nav, hero, pricing, testimonials, features,
cta, footer, sidebar, content.
Crawler emits rawData.light.sections carrying tag, role, class/id,
text sample, headings, button/card counts, and bounds for each
landmark/section element.
* feat(extractors): reusable component clustering with variant detection
Groups button/input/link/card candidates by structural hash (tag >
child tags), then splits each group into variants by cosine
similarity on a style vector (padding, bg/fg channels, radius,
border, font metrics). Cosine >= 0.95 collapses into one variant.
Crawler serializes rawData.light.componentCandidates with kind,
structuralHash, styleVector, and human-readable css per element.
Markdown formatter renders a new "Component Clusters" section
(additive — the legacy "Component Patterns" section is preserved).
* chore(repo): add issue templates, funding config, contact links
* feat(formatters): iOS SwiftUI emitter from DTCG tokens
Adds a shared internal `resolveRef(tokens, path)` helper plus
`formatIosSwiftUI(tokens)` that emits a SwiftUI file with:
- semantic Colors (resolved through {primitive.*} references)
- CGFloat extensions for primitive spacing/radius
- a primitive palette for direct use
- typography hint comments (Font generation deferred)
9 new tests (4 iOS + 5 resolveRef).
* feat(formatters): Android Compose emitter
Adds `formatAndroidCompose(tokens)` producing Theme.kt (Kotlin Compose
object with Color/dp values), plus colors.xml and dimens.xml resources.
PascalCase in Kotlin, snake_case in XML, #FF-prefixed hex for alpha.
3 new tests.
* feat(formatters): Flutter Dart emitter
Adds `formatFlutterDart(tokens)` producing a Dart file with a
DesignTokens class (semantic colors, primitive spacing/radius/colors)
and a buildDesignlangTheme() helper wiring primary/surface into a
ColorScheme.
2 new tests.
* feat(formatters): WordPress block theme skeleton with theme.json
Adds `formatWordPressTheme(tokens, design)` returning a full block-theme
file map:
- theme.json (WP schema v3 with color palette, spacing, font sizes/families
derived from DTCG tokens)
- style.css (valid WP theme header + :root custom props)
- functions.php (minimal theme support)
- index.php (template-part fallback)
- templates/index.html (block-markup entry)
Legacy `formatWordPress(design)` export preserved for existing consumers.
4 new tests.
* feat: --platforms dispatcher for web/ios/android/flutter/wordpress
Adds a comma-separated --platforms CLI option that emits additional
platform-native outputs alongside the existing web set:
- ios → ios/DesignTokens.swift
- android → android/Theme.kt + colors.xml + dimens.xml
- flutter → flutter/design_tokens.dart
- wordpress → wordpress-theme/{theme.json, style.css, functions.php,
index.php, templates/index.html}
"all" expands to every known platform. Web is always emitted (additive).
Threads `platforms` through mergeConfig with a parsePlatforms helper that
normalizes CSV/array inputs, ignores unknown values, and guarantees web
is present.
10 new tests (6 parsePlatforms, 3 mergeConfig platforms, 1 CLI help).
Manav Arya Singh