Hallmark by Nutlope: Anti-AI-Slop Design Skill for Claude Code, Cursor, and Codex
Nutlope/hallmark (11.1k GitHub stars) is a Together AI design skill with 20 themes, 57 slop-test gates, and four verbs — build, audit, redesign, study. Install via explainx.ai/skills or npx skills add.
If you've asked an agent to "build a landing page" and gotten the same purple gradient, three identical feature cards, and a hero that says "Revolutionize your workflow" — you've seen the problem Hallmark targets. Nutlope/hallmark is an anti-AI-slop design skill for Claude Code, Cursor, and Codex: it encodes rules that refuse the on-distribution defaults every frontier model was trained on, then runs 57 slop-test gates before handing code back.
Made by Together AI, Hallmark reached 11.1k GitHub stars and 553 forks by July 2026. Live examples cycle at usehallmark.com — press T to swap themes. On explainx.ai, install from the Hallmark skill page (Cursor guide, agent matrix, and one-line CLI copy included).
Self-contained HTML + CSS pages, or in-place edits respecting existing tokens
Why another design skill?
Most agent UIs fail on structure, not just colour. Two briefs — a sourdough app and a record-label EP — should not share the same hero → three features → CTA → footer rhythm. Hallmark's differentiator (documented in references/structure.md) is macrostructure variety: it picks layout archetypes per brief, dresses them in one of twenty named themes (Hum, Cobalt, Carnival, Lumen, Garden, Riso, modern-minimal, atmospheric, etc.), and only falls back to Custom when the brief carries creative intent no catalog theme fits.
The skill draws from the consensus anti-slop field — Anthropic's frontend-design guidance, Claude cookbook aesthetics, and the 2026 "tactile rebellion" movement — but packages it as enforceable gates, not vibes.
Four verbs
Verb
What it does
(default)
Build new UI. Pre-flight scan → genre → theme → macrostructure → slop test → emit.
hallmark audit <target>
Score existing code against anti-patterns. Ranked punch list; no edits.
Select Cursor (or your agent) when the CLI prompts
Verify: .cursor/skills/hallmark/ exists · restart Cursor · invoke via /hallmark
The registry page includes security notes (review SKILL.md source before production), use cases, and links to related design skills. Always review skill source — skills execute in your environment.
Manual paths (from README)
Agent
Path
Claude Code
~/.claude/skills/hallmark/
Cursor
.cursor/skills/hallmark/
Codex
~/.codex/skills/hallmark/ or .codex/skills/hallmark/ (project)
Copy SKILL.md + references/ from the repo, or re-run npx skills add to update.
What good output looks like
Hallmark's README gallery shows brief-specific shapes — not colour-swaps of one template:
Example
Brief
Theme
Bubble
Guided sourdough app
Hum
Distil
Content-extraction API
Cobalt
Cold Snap
Record-label EP
Carnival
Tally
SaaS product page
modern-minimal
Wayfare
Travel booking
atmospheric
The Mend Assembly
Repair-café broadsheet
Custom
Each page is self-contained HTML + CSS with macrostructure stamped in a CSS comment. Browse live at usehallmark.com or under site/_tests/ in the repo.
Hallmark vs detecting slop after the fact
Hallmark is generative guardrails — it stops slop at emit time. That's complementary to detection tools (classical ML classifiers, Pangram, browser extensions) that score text after publication. If you're fighting AI-written fanfic or blog spam, see explainx.ai's LLM text detection guide. If you're fighting visual sameness in agent-built UIs, Hallmark is the upstream fix.
The shadcn creativity debate and Thoughtworks zero-cost fallacy cover why cheap copying compresses lead time — Hallmark doesn't solve attribution, but it does raise the floor on what "shipped" looks like when you still care about craft.
Limits and honest trade-offs
Closed-weight models still interpret rules — Hallmark is long; context limits and model drift can skip gates. Run hallmark audit on output.
Existing codebases need pre-flight — Without design.md, first run may conflict with hard-coded font-family: Inter vs next/font Geist. Hallmark flags conflicts; you confirm.
Component vs page ambiguity — "Design a pricing section" triggers a one-question disambiguation; default is component if you don't answer.
Study mode can't judge rhythm from URL alone — URL fetch names exact fonts and colours but not spacing feel; screenshot fallback when auth-walled.
Hallmark version, star count, and theme catalog reflect the Nutlope/hallmark README and explainx.ai registry listing as of July 17, 2026. Re-run npx skills add to pull updates; verify SKILL.md before production installs.