Hallmark Skill Promises AI‑Generated Design Detection for Claude Code, Cursor, Codex

Overview

A new open‑source design skill called Hallmark has been released on GitHub. It is marketed as a tool that can *refuse to look AI‑generated* and flag common “sloppy” patterns that often appear in automatically‑produced web pages. The skill can be attached to Claude Code, Cursor, Codex and OpenCode, allowing developers to run a lint‑like check before a site goes live.

What Hallmark Claims to Do

• Detects AI‑style cues such as generic eyebrow headings, tracked‑capital labels, single‑colored words in headlines, empty frames, hero images that miss the viewport, and placeholder email sign‑ups.
• Offers a catalog of 50 “themes.” Each theme is a material (paper, type, colour, texture) with a reference screenshot. The build process selects one theme that matches the project’s purpose (booking page, shop, newsletter, etc.).
• Runs a “fresh‑context reviewer.” After the page is rendered in a real browser, a sub‑agent examines eight viewport widths and reports any detected slop.
• Supports custom prompts. Users can ask Hallmark to generate a page from a natural‑language prompt alone, while still applying the same anti‑slop checks.
• Provides redesign and audit modes. `/hallmark redesign` rebuilds a page keeping its content and wiring; `/hallmark audit` only reports issues.

Confirmed Technical Details (from the repository README)

• The skill is distributed as an npm package; it requires Node 18+ and, for full browser checks, puppeteer‑core and a Chrome binary.
• All generated pages are self‑contained HTML + CSS files, with a comment block that records the decisions made during the build.
• Hallmark uses a library of 755 openly‑licensed photographs sourced from Wikimedia Commons. Licences (CC0, CC BY, CC BY‑SA) are credited on the page.
• Installation commands include:

``` npx skills add nutlope/hallmark -g --agent claude-code ``` or adding the plugin via Claude Code’s marketplace.

• The lint script `sloplint.mjs` can be run manually (`node scripts/sloplint.mjs --render`) to see the eight‑width render report.

Practical Implications for Developers

• Quality gate: Teams that rely on AI‑assisted code generation can add Hallmark as an automated gate to catch visual patterns that often betray machine output.
• Design consistency: By selecting from a fixed set of themes, projects can maintain a coherent visual language without hand‑crafting every style.
• Rapid prototyping: The custom‑prompt mode lets a developer describe the desired page (e.g., “a booking page for yoga classes”) and receive a ready‑to‑use HTML file that respects the anti‑slop rules.
• Audit trail: The embedded comment with decision metadata offers a lightweight audit log for compliance or design review.

How to Verify the Claims

1. Clone the repo (`git clone https://github.com/nutlope/hallmark`)
2. Run the test suite (`npm test` if provided) to see lint checks in action.
3. Generate a sample page using the default theme and inspect the HTML comment for decision stamps.
4. Compare before/after with `/hallmark redesign` on an existing site to confirm that content is preserved while layout changes.
5. Check photo credits on the generated page to ensure the Wikimedia licences are displayed as described.

Unknowns / Open Questions

• The README does not specify performance metrics (e.g., how long a full eight‑width render takes on typical hardware).
• It is unclear how the “fresh‑context reviewer” is implemented—whether it is a rule‑based engine or a small language model.
• No public user studies or third‑party audits have been published to validate the effectiveness of the anti‑slop detection.

Getting Started

Developers can install Hallmark globally or per‑project via npm, then invoke it through the Claude Code, Cursor, Codex or OpenCode agents. The tool works on local dev servers, URLs, or static files, making it flexible for both early‑stage prototypes and production builds.

--- *This article is based on the publicly available GitHub repository for the Hallmark skill. No independent testing has been performed.*