Skip to content

Designesy Design System Contract v0.4.0

Choose a tag to compare

@LE-VAI LE-VAI released this 31 Jul 07:03
· 237 commits to main since this release

Designesy Design System Contract v0.4.0

The fourth tagged release of the Designesy design system contract — now with 40 verification checks across 14 weighted categories, including the new Copywriting adoption.

What's new in v0.4.0

Copywriting adoption (the big one)

16 UX copy principles + 4 verification checks:

  • v38 — Button text is a verb phrase or recognized command (not a bare noun)
  • v39 — No trailing period on button text, labels, or tab text
  • v40 — Descriptive link text (no "click here")
  • v41 — No ALL CAPS in body text

Grounded in NN/g guidance. Heuristic WARNs for manual review where deterministic checks can't fully cover intent.

Check count: 23 → 40

17 new checks across the existing categories plus the new Copywriting category. The verification engine now covers 14 weighted categories: tokens, motion, accessibility, cadence, takt, poise, identity, interaction, performance, responsive, semantic, security, spec, copywriting.

Originality + anti-slop signals

Positive craft signals panel in the score report — surface originality lift (+8 cap) that recognizes deliberate craft choices (glassmorphism done well, scoped gradients, intentional motion) rather than penalizing anything that looks like a pattern.

GitHub Action + CLI (adoption surfaces)

  • GitHub Actionuses: LE-VAI/designesy-org@v1, node24, PR comments, 4 emission formats
  • designesy-score CLInpx designesy-score <url>, zero-dep, npm published
  • MCP server — 11 tools, Streamable HTTP, no auth

Looping hero demo

The README and npm listing now include a looping Score Gate hero animation — a URL flows in, the 40-check contract grid fires, the score counts up to 99.2% grade A, and the gate passes. Built as a HyperFrames composition, rendered to mp4, encoded to a sub-1MB GIF with ffmpeg (24fps, 64-color palette, lanczos + bayer dithering). A static PNG fallback is served via <picture> art-direction when the OS has reduced-motion enabled.

MCP tool descriptions rewritten for agent selection (TDQS Tier-A)

All 11 MCP tool descriptions restructured against the Glama TDQS (Tool Definition Quality Score) six-dimension rubric. Research (arXiv 2602.18914) found smell-free descriptions win 72% of head-to-head agent selections vs a 20% random baseline — a 260% relative lift. Every tool now has: verb+object purpose lead, "When NOT to use" sibling disambiguation (89% of MCP tools omit this), behavioral transparency (read-only/executable/cache/SKIP disclosure), parameter intent, and return JSON shapes. Server-level TDQS = 0.6×mean + 0.4×min — all 11 must pass Tier A (≥3.5) or the worst sinks the server.

Leaderboard

30 sites scored, with an open submission form at /leaderboard — submit a URL, get scored on-the-fly, curated into the next weekly batch.

Full adoption history

Version Date Summary
0.1.0 2026-07-12 First public contract — tokens, motion, components, verification
0.1.1 2026-07-12 Poise interaction rules — wordmark breath, press settle, sound preference, reduced motion
0.1.2 2026-07-13 Takt interface-feel rules — concentric radii, press scale, image outlines, hit area floor
0.1.3 2026-07-13 Cadence typography rules — font smoothing, rem scale, line-height by role, text-wrap, tabular nums
0.1.4 2026-07-13 Duration scale, Core Web Vitals + contrast verification, anti-rationalization methodology
0.3.0 2026-07-15 Spring physics, full acoustic enumeration, 10 Motion Standards, motion anti-patterns, entrance scale tokens
0.4.0 2026-07-31 Copywriting adoption (16 principles + 4 checks), 40-check engine, originality signals, GitHub Action v1, CLI, leaderboard, looping hero demo, TDQS Tier-A tool descriptions

Machine export

Live score

The designesy.org site is verified against this contract: 99.2% (Grade A) — 35 pass, 0 fail, 1 warn, 4 skip (browser-only checks).