Hallmark Keeps Claude Code and Cursor From Making 'AI Landing Pages': 50 Themes and a Browser Linter
Landing pages from Claude Code and Cursor are easy to spot: tiny all-caps labels above every heading, one word in the headline set in a different color, and a newsletter signup form at the bottom. The Hallmark skill from Nutlope offers three themes from a catalog of 50 'materials' (paper, typeface, color, texture) for your request, and a linter checks the finished page in a real browser at eight screen widths. If your agent churns out identical websites, install the skill and compare the result with your usual output.
AI-processed from GitHub Trending AI; edited by Hamidun News
If you ask Claude Code or Cursor to build a landing page, the result is often predictable: tiny all-caps labels above every heading, one word in the headline set in a different color, a row of empty cards, and a newsletter signup form at the bottom. Such pages look 'machine-made,' and clients notice. The Nutlope/hallmark repository gained 2,859 GitHub stars in a single day (the stars function as the repository's bookmarks or 'likes').
The idea is simple: instead of asking the agent to 'make it look nice,' you put it within strict boundaries. For each request, the skill picks three themes from a catalog of 50 options and checks the finished page in a real browser at eight screen widths, from phone to wide monitor.
What Hallmark Is and How It Works
Hallmark is a skill for Claude Code, Codex, Cursor, and OpenCode. It enforces a strict minimum of quality and otherwise stays out of the way. First, the agent reads the project: fonts, palette, framework, and any decisions already locked in. Then it asks you once for materials: photos, screenshots, real figures. Nothing invented ends up on the page.
Next, the skill offers three themes from the catalog of 50 that fit your request. Here, a theme is a 'material': paper, typeface, color, surfaces, custom components, and texture. It does not dictate what goes where on the page or in what order, so two different requests in the same theme produce two different pages. The agent derives the page's structure from its purpose. A booking page leads to a reservation, a store sells, a newsletter collects subscribers, and the page ends with that action.
There is also a mode without themes. If you ask for a custom design, Hallmark keeps only the base minimum, the checks, and a short direction: find one idea that belongs to this client alone and build the page around it.
The Linter Versus 'AI Tells'
The core of the tool is result checking. The script scripts/sloplint.mjs, run with the --render flag, opens the page in a real browser and inspects it at eight screen widths. It looks for common markers of machine-generated markup: a label above every heading, letterspaced all-caps captions, a single colored word in a headline, a page made of empty boxes or nothing but photos, a first screen that doesn't fit the viewport, a signup form at the end of every page, and a form that says 'thank you' but sends nothing.
The rules fall into three tiers. Hard rules block the build: contrast, focus, reduced motion, invented facts, and broken mobile layout. The second tier covers habits the model falls into by default. These can be broken if the build has a reason, and the reason gets recorded. The third tier is minor craft notes that block nothing.
The workflow is unusual too. The first screen (hero) is built separately, rendered, and shown to the reviewer 'cold,' before the rest of the page exists. At the end, a subagent with fresh context reviews the finished result, and it has the final word. Each page consists of self-contained HTML and CSS, and the first CSS comment records the decisions made.
Redesigns, Photos, and Installation
Hallmark does not touch an existing design unless you explicitly name it. The command /hallmark redesign ./src/app rebuilds the system and markup but preserves content and logic. Before starting, the skill records every link, field, handler, and heading. If anything goes missing after the redesign, the check counts as failed. The output is a before-and-after comparison. The command /hallmark audit ./index.html only checks the file and rewrites nothing.
For images, the tool is connected to a list of 755 photos from Wikimedia Commons. The licenses vary: CC0, public domain, CC BY, or CC BY-SA, and every photo used is credited on the page. The author does not use stock sites such as Unsplash, Pexels, and Pixabay, because their terms prohibit selecting images with AI tools.
Install the skill with npx skills add nutlope/hallmark -g. It requires Node 18 or newer. For Claude Code, there is a plugin-marketplace option. Without Chrome and browser checks, only the static checks work. The license is MIT.
What to Do With It
The limitations are obvious. A catalog of 50 themes is still finite, and the linter catches only known tells: a site that passes every rule can still be dull. But the idea is useful: instead of persuading the agent in a prompt, you give it measurable criteria and an independent reviewer. If you build landing pages and prototypes with AI agents, it is worth running at least one of your own projects through Hallmark and comparing the result with your usual output.
- Nov 21, 2026Anthropic Sets IPO Price at $83 per Share, Offering Around $81 Billion
- Nov 21, 2026Anthropic to Unveil Claude for Factories: 17 Ready-Made Skills and 4 Integrations with Shop-Floor Software
- Nov 21, 2026AI Coding Assistant Under $10: Cursor at $8, Xiaomi MiMo at $6 a Month
- Nov 21, 2026Anthropic goes public and raises a record $107.3 billion
Key Questions Answered
What Hallmark Is and How It Works?
Hallmark is a skill for Claude Code, Codex, Cursor, and OpenCode. It enforces a strict minimum of quality and otherwise stays out of the way. First, the agent reads the project: fonts, palette, framework, and any decisions already locked in. Then it asks you once for materials: photos, screenshots, real figures. Nothing invented ends up on the page. Next, the skill offers three themes from the catalog of 50 that fit your request.
What are the key details on The Linter Versus 'AI Tells'?
The core of the tool is result checking. The script scripts/sloplint.mjs, run with the --render flag, opens the page in a real browser and inspects it at eight screen widths. It looks for common markers of machine-generated markup: a label above every heading, letterspaced all-caps captions, a single colored word in a headline, a page made of empty boxes or nothing but photos, a first screen that doesn't fit the viewport, a signup form at the end of every page, and a form that says 'thank you' but sends nothing.
What are the key details on Redesigns, Photos, and Installation?
Hallmark does not touch an existing design unless you explicitly name it. The command /hallmark redesign ./src/app rebuilds the system and markup but preserves content and logic. Before starting, the skill records every link, field, handler, and heading. If anything goes missing after the redesign, the check counts as failed. The output is a before-and-after comparison. The command /hallmark audit ./index.html only checks the file and rewrites nothing.
Build & Ship Production AI Agents Without Boilerplate Code
Start your free 72-hour trial at Hamidun Academy (academy.hamidun.com) — interactive lessons on Claude Code & Cursor, live Tech Radar, and production agent templates.
The AI world, distilled — once a week
Seven stories that actually mattered, hand-picked. No noise, no reposts, no press releases.
Done! Check your inbox for a confirmation.