dent
Open-source AI design skill for Claude Code, Codex, Cursor & Copilot: asks the right questions, follows UX laws, accessibility and Google's Modern Web Guidance, verifies and scores its own UI. 41 styles, 31 commands.
Safety scan
A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.
Badge for your README
Markdown
[](/listings/jeelnadaa--dent)HTML
<a href="/listings/jeelnadaa--dent"><img src="/api/listings/jeelnadaa--dent/badge.svg" alt="findsafeskills safety rating" /></a>Install
Install instructions from the README
As written by the repo's author. findsafeskills has not verified these steps.
**Claude Code (plugin, gets all 32 slash commands)**
/plugin marketplace add jeelnadaa/DENT
/plugin install dent@dent
**Any agent, one command** (detects Claude Code, Codex, Cursor, Copilot, Gemini CLI, OpenCode…)
npx skills add jeelnadaa/DENT
**Manual:** copy `skills/dent/` and `skills/modern-web-guidance/` side by side into your agent's skills folder:
| Agent | Personal (all projects) | Project only | |---|---|---| | Claude Code | `~/.claude/skills/dent/` | `.claude/skills/dent/` | | OpenAI Codex | `~/.codex/skills/dent/` | `.agents/skills/dent/` | | GitHub Copilot | `~/.copilot/skills/dent/` | `.github/skills/dent/` | | Cursor | `~/.cursor/skills/dent/` | `.cursor/skills/dent/` | | Gemini CLI | `~/.gemini/skills/dent/` | `.gemini/skills/dent/` | | Antigravity, OpenCode, Goose, Amp & other Agent Skills tools | — | `.agents/skills/dent/` |
Or run the installer (installs both): `./scripts/install.sh claude` (macOS/Linux) · `.\scripts\install.ps1 -Agent claude` (Windows).
**Claude.ai / Claude Desktop:** download `dent.zip` from [Releases](https://github.com/jeelnadaa/DENT/releases) and upload it under **Settings → Capabilities → Skills** (it includes Modern Web Guidance; upload `modern-web-guidance.zip` as well if you want it active outside DENT). Agents without skill support (Windsurf, Cline, Aider, Zed…): see [docs/INSTALL.md](docs/INSTALL.md#agents-without-native-skill-support).
**Updating:** run `/dent:update` in Claude Code, then click **Update** on DENT in Settings → Plugins (or `/plugin update dent@dent`). Turn on auto-update for the DENT marketplace to get new versions automatically. Details: [docs/INSTALL.md](docs/INSTALL.md#updating).
**Optional, for real verification:** `pip install playwright && python -m playwright install chromium` lets DENT run `verify_page.py` on what it builds. Node.js lets it use the always-current Modern Web Guidance CLI (`npx`); without it DENT reads the bundled guides.
> Skill folder paths change as tools evolve. If a path doesn't work, check your agent's docs for its "skills" directory. Full details: [docs/INSTALL.md](docs/INSTALL.md).
---
What was scanned
The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.
.claude-plugin/marketplace.json · Manifest · 26 lines
- {
- "name": "dent",
- "owner": {
- "name": "Jeel Nada (solarquack)",
- "url": "https://github.com/jeelnadaa"
- },
- "metadata": {
- "description": "DENT: open-source product design skill for AI agents.",
- "version": "2.4.0"
- },
- "plugins": [
- {
- "name": "dent",
- "source": "./",
- "description": "Design-engineering skill: 32 commands, orchestration, planning + plan audit, 41 styles + directions library, component/motion/audio/video catalogues, verify_page.py, self-scoring and DESIGN.md. Includes Google's Modern Web Guidance skill as the web-platform authority.",
- "version": "2.4.0",
- "author": {
- "name": "Jeel Nada (solarquack)"
- },
- "homepage": "https://github.com/jeelnadaa/DENT",
- "license": "Apache-2.0",
- "category": "design",
- "keywords": ["design", "ui", "ux", "design-system", "accessibility"]
- }
- ]
- }
README.md · README · 446 lines
- <a href="#install">
- <picture>
- <source media="(prefers-color-scheme: dark)" srcset="assets/hero-dark.svg">
- <img src="assets/hero-light.svg" alt="DENT 2.0: a generic AI-template layout is knocked away by an orange dot and replaced by the DENT wordmark and the line 'Make a dent with your product's design.'" width="100%">
- </picture>
- </a>
- # DENT: the AI design skill for Claude Code, Codex, Cursor and Copilot
- **DENT is an open-source design-engineering skill for AI agents.** Install it once and your agent (Claude Code, Codex, Cursor, Copilot, Gemini CLI, Antigravity, or anything that reads `SKILL.md`) designs like a senior product designer, builds like a front-end engineer, animates like a motion designer, and scores like a strict reviewer. Interfaces, motion graphics, 3D, sound, narration and video.
- [](LICENSE)
- [](https://agentskills.io)
- [](CHANGELOG.md)
- [](#modern-web-guidance-the-authority-for-web-code)
- [](skills/dent/references/styles/_styles-index.md)
- [](docs/COMMANDS.md)
- **[Website](https://dent.solarquack.in)** · **[Install](#install)** · **[One command](#one-command-the-right-pipeline)** · **[What it can make](#what-dent-can-make)** · **[Proof](#see-the-difference-run-the-benchmark)** · **[Commands](docs/COMMANDS.md)** · **[Architecture](docs/ARCHITECTURE.md)**
- ---
- ## The problem
- Ask any agent for a landing page and you get the average of every template: a violet gradient, a glowing orb, a centered headline, three equal cards, invented "loved by 10,000+ teams" stats, and a phone layout that scrolls sideways. It looks finished for a second and forgettable forever.
- **DENT changes how the agent works, not just what it outputs.**
- | Without DENT | With DENT 2.0 |
- |---|---|
- | Guesses the look from one sentence | Classifies the task, asks the few questions that matter, or decides for you and shows why |
- | The category cliché | Names the cliché, then diverges: concept, one of 41 styles, a directions library, a signature hero |
- | Invents social proof | **Never.** Honesty is a gate: real proof or clearly marked placeholders |
- | "Looks fine on my screen" | `verify_page.py` checks overflow at 6 widths, contrast, labels, reduced motion, autoplay audio |
- | Adds a second UI library to your repo | Detects your design system and extends it; picks libraries from sourced catalogues, licences included |
- | Writes yesterday's web code: JS modals with focus traps, positioning libraries, global animation kill switches | Follows Google's **Modern Web Guidance** first: native `<dialog>`, `popover`, anchor positioning, `<details>`, view transitions, with each guide's fallbacks |
- | Leaves OS-default tooltips, scrollbars, sliders and checkboxes on a designed page | Asks *match my design / brand tint / native*, then styles every built-in surface with your tokens and checks it (`/dent:surfaces`) |
- | Fades everything in | Motion with a purpose, storyboards for motion graphics, reduced-motion versions |
- | Click sounds everywhere, autoplaying music | Sound as a designed, opt-in system; narration with captions |
- | "Done!" | Scores itself against *your* request with evidence; `/dent:improve` raises the score |
- ---
- ## One command. The right pipeline.
- <picture>
- <source media="(prefers-color-scheme: dark)" srcset="assets/workflow-dark.svg">
- <img src="assets/workflow-light.svg" alt="Diagram: any request goes to /dent, which builds a pipeline (route, plan, audit, direct, build, verify, score, improve) powered by knowledge, catalogues, quality checks and memory." width="100%">
- </picture>
- You never have to learn the commands. Type `/dent` (or just describe what you want) and DENT's router works out the deliverable, the state of your project, the media involved and the risk, then chains the right sub-commands:
- ```text
- /dent go all out: a landing page for my app and a 45-second launch video with voice-over
- ```
- → *two pipelines, one shared design system:* `directions → create → type → motion-graphics → assets → verify → score → improve`, then `launch-video → voice → sound → verify → score`.
- ```text
- /dent this dashboard scrolls sideways on my phone
- ```
- → `responsive → verify`. No questionnaire for a bug.
- Power users can call any of the [32 commands](docs/COMMANDS.md) directly: `/dent:plan`, `/dent:audit`, `/dent:components`, `/dent:motion-graphics`, `/dent:sound`, `/dent:video`, `/dent:score`, `/dent:improve`…
- ---
- ## What DENT can make
- | Discipline | What DENT brings |
- |---|---|
- | **Product design** | Six modes (create, redesign, audit, extend, implement, adopt), 4 effort levels, 4 task depths, UX laws labelled Law / Rule / Convention, recipes for landing pages, apps and dashboards |
- | **Creative direction** | Concept sentence, category-cliché check, 41 aesthetic styles, a [library of 25 named directions](skills/dent/references/creative/directions-library.md) plus motion, sound and video directions, a [hero typography system](skills/dent/references/creative/hero-typography.md) with 18 composition techniques |
- | **Front-end engineering** | Tokens first, ~45 component specs, code guides for HTML/CSS, Tailwind, React, Vue, Svelte, Flutter, SwiftUI, Compose, React Native. On the web, [Modern Web Guidance](#modern-web-guidance-the-authority-for-web-code) decides how each piece is built |
- | **Component intelligence** | A [sourced catalogue](skills/dent/references/catalog/README.md) of primitives (Radix, Base UI, React Aria, Ark…), systems (MUI, Mantine, Ant, Carbon, Fluent…), collections (shadcn/ui, Magic UI, Aceternity…) and specialists (TanStack Table, AG Grid, Tiptap, cmdk, Sonner, dnd-kit, React Flow…), with licences and paid tiers called out |
- | **Motion graphics** | Interface motion vs motion graphics vs 3D, 1D → 2D → 2.5D → 3D → XR, storyboards, kinetic type, scroll stories, generative visuals, and a tech ladder from CSS to GSAP, Rive, PixiJS, Three.js and shaders |
- | **Sound and voice** | UI sound, sonic identity, ambience, interactive and procedural audio, music, narration and TTS, captions; never autoplay, always opt-in |
- | **Video** | Launch (teaser, reveal, feature tour, manifesto), trailers, explainers, demos, social, ads; scripts, storyboards, animatics; generative tools if you have them, otherwise HyperFrames, Remotion, Motion Canvas and FFmpeg |
- | **Quality** | 13 gates, runnable checks, an [All-out protocol](skills/dent/references/workflow/all-out.md) that misses nothing, request-specific self-scoring, and attribution of what came from DENT vs the agent vs you |
- ---
- ## Modern Web Guidance: the authority for web code
- DENT bundles Google Chrome's [Modern Web Guidance](https://developer.chrome.com/docs/modern-web-guidance) skill (stewarded with the Edge, Firefox and MDN teams, Apache-2.0) and makes it the **primary authority for every HTML, CSS and client-side JS task**. DENT decides *what* is designed; MWG decides *how* it is built on the web.
- - **Before writing web code** DENT searches and retrieves the relevant guides (`npx -y modern-web-guidance@latest search …`, or offline with `skills/dent/scripts/mwg.py`), applies them, checks the result against them (gate G13) and ends with a short `### Modern Web Guidance` note naming the guides it followed.
- - **MWG wins conflicts.** DENT 2.3 was audited against it: global reduced-motion resets, hand-made modal focus traps, split OTP inputs, library-first widgets and oversized fluid-type ranges were replaced with MWG's patterns ([full table](skills/dent/references/modern-web-guidance.md#6-conflicts-resolved-in-dent-230)).
- - **Native first:** `<dialog>`, `popover`, anchor positioning, `<details>`, `<input switch>`, customizable `<select>` before any library; libraries only where the platform has no primitive.
- - `verify_page.py` flags the mechanical violations (a global `*` reduced-motion reset, a lazy-loaded LCP image, ARIA re-added to `<dialog>`, split one-time-code inputs).
- The plugin ships MWG as its own skill too (`skills/modern-web-guidance/`, vendored verbatim and pinned in `UPSTREAM.json`), so it also kicks in on web tasks that never touch DENT.
- ---
- ## Built with DENT 2.0
- The next showcase pieces are being produced **with DENT 2.0 itself**, using the exact prompts in [`evals/scenarios.md`](evals/scenarios.md) (S19–S21). Every question the agent asks and every answer given will be published with the result.
- | Piece | Status |
- |---|---|
- | DENT's own landing page | Next |
- | A kinetic-typography motion piece about DENT | Next |
- | A 60–75 s explainer video with narration and captions | Next |
- The animated banner at the top of this page is hand-built SVG: the wordmark is drawn as paths, it plays once, and it respects reduced-motion settings. Its source is [`assets/build_assets.py`](assets/build_assets.py).
- ---
- ## See the difference: run the benchmark
- Don't take our word for it. Give your agent the **same prompt twice**, once without DENT and once with it, and compare the two websites. We publish every result: wins, ties and losses.
- **How to run either benchmark**
- 1. Use the same agent, model and settings for both runs, each in a **fresh session** and an **empty folder**.
- 2. **Run A, without DENT:** skill not installed. **Run B, with DENT:** skill installed ([install](#install)).
- 3. Paste the prompt unchanged. Apart from answering the agent's questions (benchmark 1 only), reply only `continue` if it stops early. Keep the first result of each run.
- 4. Open both `index.html` files side by side on your phone and desktop, and share your screenshots via a pull request or a [Benchmark result issue](https://github.com/jeelnadaa/DENT/issues/new?template=benchmark_result.md).
- Full rules, scoring sheet and a screenshot script: **[benchmark/](benchmark/README.md)** · Full test protocol: **[evals/protocol.md](evals/protocol.md)**
- ### Latest results: 8 builds with and without DENT (Claude Opus 5.5)
- Same agent (Claude Code), same model (Claude Opus 5.5, medium effort), same prompt, built twice: once without DENT and once with DENT 2.0. Full summary: **[RESULTS-SUMMARY.md](benchmark/RESULTS-SUMMARY.md)**.
- | Test | Without DENT | With DENT | Comparable? |
- |---|---|---|---|
- | [Brewlog marketing site](benchmark/results/brewlog-claude-code-opus-5.5-medium-2026-11-14/RESULT.md) | 38 | **44** | Yes |
- | [SaaS analytics dashboard](benchmark/results/saas-dashboard-claude-code-opus-5.5-medium-2026-11-14/RESULT.md) | 39 | **43** | Yes |
- | [Checkout with failure recovery](benchmark/results/checkout-flow-claude-code-opus-5.5-medium-2026-11-14/RESULT.md) | 39 | **44** | Partly |
- | [Kanban + command palette + drawer](benchmark/results/project-app-claude-code-opus-5.5-medium-2026-11-14/RESULT.md) | 37 | **43** | Partly |
- | [Scroll-story launch page](benchmark/results/scroll-story-launch-claude-code-opus-5.5-medium-2026-11-14/RESULT.md) | 37 | **45** | Partly |
- | [Meditation timer with sound](benchmark/results/meditation-sound-claude-code-opus-5.5-medium-2026-11-14/RESULT.md) | 39 | **44** | Partly |
- | [3D product viewer](benchmark/results/3d-product-viewer-claude-code-opus-5.5-medium-2026-11-14/RESULT.md) | 36 | **44** | Partly |
- | [45 s launch video](benchmark/results/launch-video-claude-code-opus-5.5-medium-2026-11-14/RESULT.md) | 38 | **43** | Partly |
- | **Average (/50)** | **37.9** | **43.8** | |
- | | Without DENT | With DENT |
- |---|---|---|
- | Scroll-story launch page |  |  |
- | 3D product viewer |  |  |
- **What the numbers say, honestly:**
- - **Automated checks:** all 7 web builds with DENT passed every automated gate. All 7 without DENT failed at least one: text contrast in all 7, animation that ignores reduced motion in 3, an unlabeled input in 1.
- - **Honesty:** one build without DENT invented "4.8 · 312 reviews". None of the DENT builds invented proof.
- - **Why the gap isn't bigger:** Opus 5.5 is a very strong model, and its pages are good on their own. DENT's lift shows most in correctness, distinctiveness (a concept instead of the template look) and process (plans, specs, self-checks). Expect a bigger gap with smaller models.
- - **Caveats:** the scores come from Claude Opus 5.5, the same model family and DENT's author, so they are not independent. Six of the eight DENT runs got a higher effort level, design feedback or extra features; every difference is listed in [DEVIATIONS](benchmark/results/DEVIATIONS-claude-code-opus-5.5-medium-2026-11-14.md). Human reviews are welcome.
- ### Benchmark 1: DENT's own landing page, built with DENT
- We are building DENT's landing page with DENT itself. Here is the exact prompt, so you can build it **with and without DENT** and compare.
- This prompt gives the agent the **content** (headlines, copy, links) and nothing about the look. It also **lets the agent ask questions**, the way real requests work: answer them briefly, and give the same answers in both runs where the questions overlap.
- ```text
- Build the landing page for DENT, an open-source product design skill for AI coding agents.
- Put it in a single self-contained index.html (HTML, CSS and JavaScript inline; Google Fonts
- allowed; no other external files, frameworks or images, so build any visuals with HTML, CSS
- or inline SVG). It must work well on phones, tablets and desktops.
- Every "Download" or "Get DENT" button links to https://github.com/jeelnadaa/DENT and opens
- in a new tab.
- Use this content. You may shorten or rephrase it lightly and choose the order of sections,
- but keep the facts accurate and don't invent statistics, testimonials, logos or user counts.
- PRODUCT
- - Name: DENT
- - Tagline: Make a dent with your product's design.
- - What it is: a free, open-source skill (a folder with a SKILL.md file) that gives AI coding
- agents the judgment of a senior product designer.
- - Works with: Claude Code, OpenAI Codex, Cursor, GitHub Copilot, Gemini CLI, OpenCode and any
- agent that supports Agent Skills (SKILL.md). Also Claude.ai and Claude Desktop.
- - License: Apache 2.0. Created by Jeel Nada (solarquack).
- HERO
- - Headline: AI agents can build anything. Now they can design it, too.
- - Subheadline: DENT is an open-source design skill that makes your AI agent ask the right
- questions, follow proven UX laws and accessibility rules, and ship interfaces that don't
- look like every other AI-made site.
- - Primary button: Download on GitHub (link above)
- - Secondary button: See how it works (scrolls to the "How it works" section)
- - Small note under the buttons: Free and open source · Apache 2.0 · Works with Claude Code,
- Codex, Cursor, Copilot and more
- THE PROBLEM
- - Heading: Every AI-made website looks the same.
- - Body: Ask an agent for a landing page and you get the usual: a purple gradient, a centered
- headline, three feature cards, one font at one weight, missing states and text you can
- barely read. It works for a second and looks like everything else forever.
- WITHOUT DENT vs WITH DENT (a comparison)
- - Guesses the look from one sentence → Asks a few focused questions, or decides for you if
- you say so
- - Default template look → A clear concept, a deliberate style and a signature element
- - One-off values everywhere → Design tokens and components reused on every screen
- - Happy path only → Empty, loading, error, success, hover, focus and disabled states
- - "Looks accessible" → WCAG 2.2 AA rules no style is allowed to break
- - Adds a second design system to your codebase → Detects your existing UI library and
- extends it
- - "I picked blue because it's nice" → A decision log with the reason behind every choice
- - Forgets everything next session → An approved DESIGN.md that becomes the source of truth
- HOW IT WORKS (5 steps)
- 1. Ask: DENT asks how far to go, then only the questions that matter, one at a time, each
- with a recommended answer.
- 2. Direct: it picks a concept and a style on purpose, not by default.
- 3. Build: tokens first, then components with every state, then screens.
- 4. Check: it reviews its own work against a quality checklist and, when the agent can,
- takes screenshots and fixes what's wrong.
- 5. Document: once you approve, everything goes into DESIGN.md so the next change stays
- consistent.
- FEATURES
- - Four effort levels: Essential, Standard, Polished and All-out. You choose the depth.
- - 41 aesthetic styles: from Minimal, Swiss and Editorial to Neo-Brutalism, Glassmorphism,
- Japandi, Dark Tech, Retro OS and Cyberpunk, each with rules for doing it well.
- - Proven principles: Gestalt, Fitts's Law, Hick's Law, Jakob's Law and Nielsen's heuristics,
- applied for you.
- - Accessibility built in: contrast, touch targets, keyboard and focus, reduced motion.
- - Works with your code: detects shadcn/ui, MUI, Chakra, Tailwind and your existing tokens,
- and extends them.
- - Design to code: HTML/CSS, Tailwind, React, Vue, Svelte, Flutter, SwiftUI, Jetpack Compose
- and React Native.
- - Six modes: create, redesign, audit, extend, implement, and adopt an existing design system.
- - Real product patterns: landing pages, pricing, dashboards, onboarding, checkout, settings,
- AI and chat interfaces, billing, roles, docs sites and more.
- STYLES SHOWCASE
- - Heading: 41 styles. Zero templates.
- - Show a sample of style names: Minimal, Swiss, Editorial, Luxury, Japandi, Neo-Brutalism,
- Glassmorphism, Dark Tech, Bento, Utilitarian, Retro OS, Y2K, Synthwave, Cyberpunk,
- Terminal, Pixel Art, plus "and 25 more".
- BENCHMARK
- - Heading: Don't take our word for it.
- - Body: We give an AI agent the same prompt twice, once without DENT and once with it, and
- publish both results: wins, ties and losses. This page itself was built with DENT. Run the
- benchmark yourself and share what you get.
- - Button: Run the benchmark (link: https://github.com/jeelnadaa/DENT/tree/main/benchmark)
- INSTALL
- - Heading: Install in one line.
- - Show these as copyable commands, each with a copy button:
- - Any agent: npx skills add jeelnadaa/DENT
- - Claude Code: /plugin marketplace add jeelnadaa/DENT then /plugin install dent@dent
- - Note: Or download the folder from GitHub and drop it into your agent's skills directory.
- Full guide: https://github.com/jeelnadaa/DENT/blob/main/docs/INSTALL.md
- FAQ
- - Is DENT free? Yes. It's open source under the Apache 2.0 license, including for
- commercial work.
- - Which agents does it work with? Any agent that supports Agent Skills (SKILL.md): Claude
- Code, Codex, Cursor, GitHub Copilot, Gemini CLI, OpenCode and more. For other agents you
- can point a rules file at it.
- - Will it slow my agent down? DENT loads one main file and opens its detailed guides only
- when a task needs them.
- - Do I own what I design with it? Yes. DENT claims no rights over your designs.
- - Can it work on my existing app? Yes. It detects your existing design system and extends
- it, or audits your current UI and ranks what to fix.
- - Can I contribute? Yes. New styles, patterns and fixes are welcome on GitHub.
- FINAL CALL TO ACTION
- - Heading: Give your agent taste.
- - Body: Install DENT in a minute and ask for your next screen.
- - Button: Download on GitHub (link above)
- FOOTER
- - DENT · Make a dent with your product's design.
- - Links: GitHub (https://github.com/jeelnadaa/DENT) · Install guide · Benchmark · License
- (https://github.com/jeelnadaa/DENT/blob/main/LICENSE)
- - Created by Jeel Nada (solarquack). Apache 2.0.
- When you are done, give me a short summary of the design decisions you made.
- ```
- **Results: DENT landing page**
- | Agent + model | Without DENT | With DENT | Run by |
- |---|---|---|---|
- | *Official results coming soon* | | | [@jeelnadaa](https://github.com/jeelnadaa) |
- Prompt file and run notes: [benchmark/PROMPT-DENT-LANDING.md](benchmark/PROMPT-DENT-LANDING.md)
- ### Benchmark 2: "Brewlog", a one-shot website
- A fictional coffee app. This prompt never mentions DENT and tells the agent not to ask questions, so both runs are one-shot and directly comparable.
- ```text
- Build a complete, production-quality marketing website for "Brewlog", a fictional app
- for home coffee brewers. Brewlog lets people log brew recipes (beans, grind size, water
- temperature, ratio, time), rate each cup, and see which settings make their best coffee
- over time. It has a free plan and a paid "Pro" plan ($6/month or $60/year) that adds
- unlimited recipes, bean inventory tracking and brew analytics.
- Requirements:
- - One self-contained file named index.html (HTML, CSS and JavaScript inline). Web fonts
- from Google Fonts are allowed. No other external files, frameworks or images: build any
- visuals (product screenshots, illustrations, icons) with HTML, CSS or inline SVG.
- - Sections: a header with navigation, hero, how it works, key features, a preview of the
- app's interface, pricing (Free vs Pro, monthly/yearly), FAQ, a sign-up form, and a footer.
- - The sign-up form asks for name, email and password, validates in the browser and shows
- helpful errors and a success state. It does not need a backend.
- - It must look great and work well on phones, tablets and desktops.
- Make every decision yourself and don't ask me any questions. When you are done, reply with
- a short summary of the design decisions you made.
- ```
- **Results: Brewlog**
- **Run 1: Antigravity + Gemini 3.8 Flash (High effort)**, 2026-11-14, run by [@jeelnadaa](https://github.com/jeelnadaa)
- | | Without DENT | With DENT |
- |---|---|---|
- | Desktop |  |  |
- | Phone |  |  |
- | Full page | [Desktop](benchmark/results/brewlog-antigravity-gemini-3.8-flash-high-2026-11-14/without-1440.jpg) · [Phone](benchmark/results/brewlog-antigravity-gemini-3.8-flash-high-2026-11-14/without-390.jpg) | [Desktop](benchmark/results/brewlog-antigravity-gemini-3.8-flash-high-2026-11-14/with-1440.jpg) · [Phone](benchmark/results/brewlog-antigravity-gemini-3.8-flash-high-2026-11-14/with-390.jpg) |
- | Live site | [Open](https://jeelnadaa.github.io/DENT/benchmark/results/brewlog-antigravity-gemini-3.8-flash-high-2026-11-14/without/index.html) | [Open](https://jeelnadaa.github.io/DENT/benchmark/results/brewlog-antigravity-gemini-3.8-flash-high-2026-11-14/with/index.html) |
- | Score (/50) | **36** | **35** |
- **Honest verdict: about a tie.** On this model both runs produced a very similar dark espresso-and-orange site. DENT clearly improved the process: a written brief and concept, a full `DESIGN.md`, reduced-motion support, a light/dark theme and clearer pricing states. But it also shipped a sideways-scrolling header on phones and kept invented stats that its own rules forbid, so the run without DENT scored slightly higher (36 vs 35; the without-DENT page also overflows at 320 px and ignores reduced motion). We're fixing both in the skill. Full scores, checks and notes: **[RESULT.md](benchmark/results/brewlog-antigravity-gemini-3.8-flash-high-2026-11-14/RESULT.md)**
- *More runs on other agents and models are welcome. One run per side isn't enough to call a winner.*
- **What DENT 2.0 changed because of this run:** every failure above is now an enforced gate (overflow, honesty, labels, style mixing, category cliché, proofreading), the [landing-page recipe](skills/dent/references/recipes/landing-page.md) walks the agent through them step by step, and `verify_page.py` catches the mechanical ones automatically (its raw output for both pages is in the result folder). The next Brewlog run will use DENT 2.0.
- ---
- ## Install
- **Claude Code (plugin, gets all 32 slash commands)**
- ```bash
- /plugin marketplace add jeelnadaa/DENT
- ```
- ```bash
- /plugin install dent@dent
- ```
- **Any agent, one command** (detects Claude Code, Codex, Cursor, Copilot, Gemini CLI, OpenCode…)
- ```bash
- npx skills add jeelnadaa/DENT
- ```
- **Manual:** copy `skills/dent/` and `skills/modern-web-guidance/` side by side into your agent's skills folder:
- | Agent | Personal (all projects) | Project only |
- |---|---|---|
- | Claude Code | `~/.claude/skills/dent/` | `.claude/skills/dent/` |
- | OpenAI Codex | `~/.codex/skills/dent/` | `.agents/skills/dent/` |
- | GitHub Copilot | `~/.copilot/skills/dent/` | `.github/skills/dent/` |
- | Cursor | `~/.cursor/skills/dent/` | `.cursor/skills/dent/` |
- | Gemini CLI | `~/.gemini/skills/dent/` | `.gemini/skills/dent/` |
- | Antigravity, OpenCode, Goose, Amp & other Agent Skills tools | — | `.agents/skills/dent/` |
- Or run the installer (installs both): `./scripts/install.sh claude` (macOS/Linux) · `.\scripts\install.ps1 -Agent claude` (Windows).
- **Claude.ai / Claude Desktop:** download `dent.zip` from [Releases](https://github.com/jeelnadaa/DENT/releases) and upload it under **Settings → Capabilities → Skills** (it includes Modern Web Guidance; upload `modern-web-guidance.zip` as well if you want it active outside DENT). Agents without skill support (Windsurf, Cline, Aider, Zed…): see [docs/INSTALL.md](docs/INSTALL.md#agents-without-native-skill-support).
- **Updating:** run `/dent:update` in Claude Code, then click **Update** on DENT in Settings → Plugins (or `/plugin update dent@dent`). Turn on auto-update for the DENT marketplace to get new versions automatically. Details: [docs/INSTALL.md](docs/INSTALL.md#updating).
- **Optional, for real verification:** `pip install playwright && python -m playwright install chromium` lets DENT run `verify_page.py` on what it builds. Node.js lets it use the always-current Modern Web Guidance CLI (`npx`); without it DENT reads the bundled guides.
- > Skill folder paths change as tools evolve. If a path doesn't work, check your agent's docs for its "skills" directory. Full details: [docs/INSTALL.md](docs/INSTALL.md).
- ---
- ## Use it
- Just ask. These all work with plain `/dent` or no command at all:
- ```text
- Design a landing page for my meditation app.
- Redesign our settings screen. Keep the brand colors.
- Review this dashboard screenshot and tell me what's wrong.
- Build the pricing page from DESIGN.md in Next.js + Tailwind.
- Add sound to the app, but keep it subtle.
- Make a 60-second explainer video with a calm narrator.
- Go all out on an onboarding flow for a fintech app. Don't ask, just decide.
- Score it. Then improve it.
- ```
- **Effort levels:** Essential · Standard · Polished · **All-out**. All-out is a promise: the best achievable result for your use case with nothing missed, a mandatory completeness sweep, and self-review until the work scores 9/10 with evidence.
- ---
- ## Architecture
- ```
- skills/dent/
- ├── SKILL.md # the router: classify → route → gates → score (lean, always loaded)
- ├── commands/ # 32 command definitions (one source of truth per command)
- ├── references/ # loaded only when needed
- │ ├── workflow/ # orchestration, planning + plan audit, execution, verification, scoring, all-out
- │ ├── recipes/ # step-by-step procedures with gates (landing page, app & dashboard)
- │ ├── creative/ # directions library, hero typography, anti-slop detector
- │ ├── catalog/ # components, primitives, specialists, assets + selection method
- │ ├── motion/ # motion graphics, tech catalogue, 3D, motion system + verification
- │ ├── audio/ # sound design, voice & narration, audio tech, implementation + verification
- │ ├── video/ # video types & craft, tools & pipeline
- │ ├── styles/ # 41 aesthetic styles
- │ └── *.md # modern-web-guidance (web authority), color, typography, accessibility, patterns, platforms, tokens & code…
- ├── templates/ # PLAN, AUDIT, REVIEW, SCORECARD, MOTION, SOUND, VIDEO
- └── scripts/ # verify_page.py, contrast.py, mwg.py (offline Modern Web Guidance lookup)
- skills/modern-web-guidance/ # Google's Modern Web Guidance skill, vendored verbatim (scripts/sync_mwg.py)
- commands/ # Claude Code plugin wrappers → /dent:<name>
- evals/ # behaviour cases, 21 scenarios, command + routing tests, fixtures
- benchmark/ # public with/without comparisons
- ```
- Progressive disclosure keeps the agent's context small: `SKILL.md` is about 330 lines; the other 130+ files load only when a task needs them. More: [docs/ARCHITECTURE.md](docs/ARCHITECTURE.md).
- ---
- ## Verification and evaluation
- - `python scripts/validate.py`: frontmatter, 500+ internal references, the styles index, the command trio (definition ↔ plugin wrapper ↔ router), 200+ catalogue entries with sources and status, versions, the vendored Modern Web Guidance copy, and every `mwg:<guide>` reference in DENT.
- - `python scripts/test_tools.py`: `verify_page.py` must fail a deliberately broken page on every gate (including each Modern Web Guidance rule it breaks) and pass a clean one.
- - [`evals/`](evals/): 15 behaviour cases, 21 scenarios (dashboards, data-heavy apps, checkout, existing design systems, unfamiliar frameworks, strict accessibility, motion graphics, 3D, animated data, audiovisual, launch and explainer videos, sonic identity) and a protocol that records the prompt, **every question asked and the answer chosen**, the output, machine checks, self-score and an independent score.
- - Catalogue licences are marked ✓ (checked against official sources on a date) or K (from knowledge, to verify), so DENT never presents a guess as a fact.
- CI runs the validator, the tool tests and the packager on every push.
- ---
- ## Roadmap
- - Publish the three showcase pieces built with DENT 2.0 (landing page, motion piece, explainer).
- - Re-run Brewlog with DENT 2.0, plus runs on more agents and models.
- - More recipes (mobile app, e-commerce, docs site, email) and more verified catalogue entries.
- - Motion verification tooling (frame-rate capture, timeline checks) alongside `verify_page.py`.
- Ideas and results are welcome: see [CONTRIBUTING.md](CONTRIBUTING.md).
- ---
- ## Creator
- DENT was created by **Jeel Nada**, known online as **solarquack** · GitHub [@jeelnadaa](https://github.com/jeelnadaa).
- If DENT helps you ship something, a ⭐ on the repo or a mention is the best thank-you.
- ## License
- DENT is licensed under the [Apache License 2.0](LICENSE). You may use, modify and distribute it, commercially too, as long as you keep the license and the [NOTICE](NOTICE) file (which credits the creator). The license does **not** grant rights to the DENT name or logo; see [TRADEMARKS.md](TRADEMARKS.md).
- Designs you create with DENT are yours. DENT claims no rights over its output.
- `skills/modern-web-guidance/` is [Modern Web Guidance](https://github.com/GoogleChrome/modern-web-guidance) by the Google Chrome team and contributors, redistributed unmodified under the Apache License 2.0 (its own `LICENSE` is included). It is not part of the DENT trademark.
Ratings and comments
Sign in to rate this skill.
Sign inFree to use. If it’s helped, pass it on.