Phrase-triggered design-system extraction. The user iterates freely on the first builds; when the system is settled, they say "lock the system" (or "give me a design.md", "make this portable") and Hallmark writes a portable ~45-line design.md at the project root. Why opt-in: - Zero churn — no design.md flickering as the user swaps themes between iterations. - The locked system reflects a settled choice, not build #1's hot take. - Mirrors how design teams actually work: formalise after patterns hold. Mechanics: - skill/references/design-md.md (new) — trigger phrases, scope rules, the tight format template, no-overwrite policy. - SKILL.md Step 3 — design-md.md added to load-conditionally. - SKILL.md Step 5 — preview block carries one quiet CTA line. Skipped on component-scope or when design.md already exists. - SKILL.md Step 6 — opt-in design.md bullet alongside tokens.css and multi-format-export rules. Default verb does NOT auto-emit. Multi- page redesign keeps producing the heavyweight design.md (unchanged). - README — one bullet under What's distinct. Page-scope only; component-scope skips. Once design.md exists every Hallmark run defers to it (existing pre-flight rule, unchanged) and the diversification rule inverts to consistency. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Hallmark
A design skill for Claude Code, Cursor, and Codex that refuses to look AI-generated.
Live demo → · twenty-three themes · four verbs · press T to cycle.
Powered by Together AI.
Hallmark encodes the anti-AI-slop consensus — typography, colour, layout, motion, microinteractions, structural variety — into one opinionated rule-set. It picks a macrostructure for the brief, dresses it in one of twenty-three themes, runs sixty-five slop-test gates plus a pre-emit self-critique before handing back, and refuses the on-distribution defaults every LLM was trained into.
v1.0.0 — pre-emit critique on six axes, honest-copy enforcement, locked-token discipline, re-drawn-chrome ban, and a no-wrap rule for clickable text. Five new slop-test gates (56–60). The full rule-set lives in
skill/SKILL.mdandskill/references/.
Try it
Paste this into Claude Code, Cursor, or Codex with the Hallmark skill installed:
"Build me a landing page for Coffeebox — a small-batch coffee subscription. Roast on Sunday, ship on Monday, drink Tuesday. Audience: people who already buy good coffee. Tone: warm, hand-set, editorial."
If Hallmark is wired correctly, you'll see Long Document · Linen (warm-paper roman-serif) · Tier-B hand-built SVG. Eight more worked briefs are documented in docs/recipes.md — copy/paste them to discover the skill before reading anything else.
What's distinct — quick map vs the field
| Hallmark | frontend-design (Anthropic) | Open Design | Dembrandt | |
|---|---|---|---|---|
| Source of taste | extracts DNA from a screenshot you admire (study verb) |
art-director brief + ban list | menu of 72 brand presets (Linear, Stripe, Vercel, Notion…) | scrapes a live URL, emits computed tokens |
| Output unit | macrostructure + theme + custom-craft | bans + brief framing | preset application | DTCG tokens.json |
| Refuses font ID | yes — names role, never guesses | n/a | n/a | n/a (computes from CSS) |
| Refuses pixel-clone | yes — DNA only, never pixels | n/a | n/a | n/a (full token export) |
| Tactile-rebellion alignment | warm-paper, custom-craft, slopless canon | strong | medium | none — token-only |
| Pages by archetype | 21 named macrostructures, picked per brief | by brief | 5 deterministic directions | n/a |
| Verbs | 4 (default · audit · redesign · study) |
1 | 31 | 1 (CLI) |
Hallmark's edge is study — every other tool ships a preset menu or a scraper. Hallmark is the only one that takes a screenshot of a design you admire, names what it sees, refuses paid-template-marketplace listings, and rebuilds your content with the extracted DNA. Three worked study examples in docs/study-examples.md.
Four verbs
| Verb | What it does |
|---|---|
| (default) | Build new UI. Asks for audience + use + tone (skippable — the skill states what it inferred). Picks a macrostructure, applies the rule-set, runs the slop test before handing back. |
hallmark audit <target> |
Score existing code against the named anti-patterns + structural sameness. Punch list, no edits. |
hallmark redesign <target> [--mood <name>] |
Throw out the structure, keep copy + IA + brand, rebuild with a deliberately different fingerprint. |
hallmark study <screenshot> |
The differentiator. Extract the DNA from a design the user admires — macrostructure, archetypes, type-pairing role, colour anchor — and produce a diagnosis report. Optionally rebuild the user's content using the extracted DNA. Refuses paid templates and competitor pages. Names font roles, never font IDs. Never copies pixels. |
Different briefs, different shapes
A handful of examples below — each generated by exercising the skill on a different brief. The skill picks differently for every prompt; no two share a macrostructure or theme.
Each page is its own self-contained HTML + CSS — no shared theme, no shared layout. Every one carries a /* Hallmark · macrostructure: … */ stamp at the top of its CSS. See the full set under site/_tests/ or live at hallmark-murex.vercel.app.
What's inside
SKILL.md— the routing file. Six-step design flow (includingStep 2.5 · Check project memoryreading.hallmark/log.json), 65-gate slop test, output contract, always-ontokens.cssexport.references/— short, opinionated rule files: typography, colour, layout, motion, microinteractions, interaction-and-states (with the input-state checklist), responsive, copy, anti-patterns, the 21 named macrostructures, the 40 component archetypes with variation knobs (9 hero · 6 feature · 4 CTA · 4 testimonial · 8 footer · 9 nav), the 4 hero polish patterns (inhero-enrichment.md), the 6 primitive structure axes, the vision-extraction protocol forstudy, hero enrichment, custom-craft (CSS art over Lottie), assets, the slop-test gates, four genre rule-overlays (each with nav + footer voice routing), per-verb dispatchers, and the export-formats reference (Tailwind / DTCG / shadcn / tokens.css).docs/— human-reading content:recipes.md(8 worked briefs + a canonical try-it prompt) andstudy-examples.md(3 worked DNA-extractions). Not auto-loaded by the skill.site/— a self-demonstrating landing page. Hand-written HTML + CSS + ES module, no framework, no build step. Twenty-three themes clustered into four genres: editorial (Specimen, Editorial, Atelier, Newsprint, Salon, Linen, Almanac, Garden, Studio, Sport, Riso, Brutal, Manifesto), modern-minimal (Quiet, Coral, Violet), atmospheric (Midnight, Terminal, Bloom, Aurora, Halo), playful (Pastel, Plume). Switching themes literally rebuilds the page — different hero archetype, different footer archetype, different nav archetype.
What's distinct (the long list)
- One skill, four verbs. Not eighteen commands.
- Genres broaden the range. Hallmark routes a brief through one of four genres before picking a theme: editorial (default · the canonical anti-slop voice), modern-minimal (Stripe / Linear / ElevenLabs school), atmospheric (Suno / Runway / dark-AI-tool school), playful (post-Linear soft school). Each genre is its own rule overlay — atmospheric allows radial blooms; modern-minimal allows pure white and pill CTAs; editorial bans both. Detection is signal-based, silent default to editorial.
- Tone is a first-class decision. "Clean and modern" is rejected. Pick an extreme — editorial · brutalist · soft · technical · luxury · playful · austere.
- Macrostructures over axes. Pick one of 21 named whole-page shapes wholesale; the macrostructure stamp lives in the CSS comment, so the next Hallmark run picks something different.
- Within-archetype variation. Two Bento Grids should not be twins; each archetype has 2–3 picked-per-output knobs.
- Microinteractions as discipline. Silent success over celebratory toasts. Optimistic update + Undo over confirm dialogs. Hover delay 800 ms, focus delay 0 ms.
- A 65-gate slop test runs before every output. One yes fails the build. Recent additions: typography discipline gates (39–40: max three font families per page, outlier face used in ≤ 2 slots), input-state gates (41–45: no border-width layout shift, focus ring via outline not border, input height matches button height, helper-text slot reserves height, disabled state needs three independent signals), contrast gates (46–50: APCA / WCAG thresholds, accent-ink token requirement, dark-section ink-on-ink check), nav · footer · hero structural gates (51–55: AI nav fingerprint, AI footer fingerprint, hero centred-everything, hero padding asymmetry, decorative-without-purpose), honest-copy + chrome + token + responsive gates (56–60: invented metrics, re-drawn UI chrome, mid-render token improvisation, two-line clickable text, emoji-as-feature-icon), and layout-safety gates (61–65: 1fr image-track overflow, missing
overflow-x: clip, display-header word-break, theme section-head mobile collapse, CSS-only radio tabs scroll-jumping). - Project memory. A per-project
.hallmark/log.jsonrecords each run's macrostructure + theme + enrichment + brief summary. The skill reads the last 3–5 entries before picking and writes a new entry after each build, so consecutive Hallmark outputs in the same project don't repeat shapes or themes. - Theme-diversification rule. Two consecutive themes must differ on at least one of three axes: paper band (dark / mid / light), display style (italic-serif / roman-serif / geometric-sans / mono / display-heavy / system-native), accent hue (warm / cool / neutral / chromatic-other).
- Voice fixtures over LLM defaults. Each of the 21 macrostructures ships with 2–3 example opening lines tuned to its tone. "Built for the modern team" is in the banned-phrases list.
- Hero enrichment is opt-in. A typographic-only hero is always acceptable. When enrichment is right, the skill picks from a six-tier hierarchy: typography only → custom-built CSS art → hand-built SVG → generated illustration (Nanobanana / Recraft) → library → Lottie (last resort).
- Microinteractions default-on for SaaS-shaped archetypes. Bento Grid, Stat-Led, Workbench, Marquee Hero pages ship with 2–3 purposeful microinteractions (number reveal, pricing lift, marquee, stagger) without the user having to ask. Editorial / Manifesto / Letter / Quote-Led pages stay still.
- SaaS page sequence. Hero → social proof → features → testimonials → pricing → FAQ → CTA → footer. Real prices, not "contact sales for pricing." Specific testimonials with role + company.
- Wordmark may use a different display face. A Geist-bodied SaaS page can set its wordmark in Fraunces. Same-family collapse on Bento / Stat-Led / Workbench / Marquee Hero is the new "un-branded" tell.
studyextracts DNA, not pixels. Refusal heuristics, type-role vocabulary (no font ID guessing), confirmation step before any code. Three worked examples indocs/study-examples.md.- Opt-in
design.mdlock-the-system flow. Iterate freely on the first builds; when the system is settled, say "lock the system" (or "give me a design.md") and Hallmark extracts the build's tokens + voice into a portable design system at the project root. From that point on every Hallmark run defers to it, the diversification rule inverts to consistency, and the file becomes the single source of truth for scaling the design across a real app. Phrase-triggered, never auto-emitted — no churn across iterations. Seeskill/references/design-md.md.
Install
npx skills add hallmark
Or copy skill/ into ~/.claude/skills/hallmark/ (Claude Code) or .cursor/rules/hallmark.mdc (Cursor — body of SKILL.md, no frontmatter).
To preview the landing page locally:
cd site && python3 -m http.server 4173 # → http://localhost:4173
Press T to cycle themes, the shuffle button (or R) to randomise, ?theme=studio for a shareable link.
Or visit the live deploy at hallmark-murex.vercel.app.
Licence
MIT. Use it, fork it, ship it.
