urmzd/dotfiles / dot_agents/skills/style-brand
style-brand
VerifiedDesign, refine or critique anything a person looks at: websites, product UI, agent workspaces, single screens and components, and brand identity. Holds the shared design standard, the mishaps to check for, and the designers we study. Use when building or changing a page or screen, reviewing a UI or a screenshot, fixing alignment, spacing, clutter or inconsistent controls, keeping several sites on one brand, or when the user shares a design or portfolio link as a reference. Also covers terminal themes and demo assets.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 6.2 KB |
| references/ai-workspace-design.md | markdown | 6.0 KB |
| references/web-design.md | markdown | 5.7 KB |
| references/who-to-study.md | markdown | 6.3 KB |
references/who-to-study.md
View on GitHub# Who to study
The designers and sites we track, and how to use them. Use this file when you
need precedent for a design decision, or when the user shares a design,
portfolio or product link.
## How to study a reference
A reference is something to learn from, not a template. In one past session the
agent opened a shared link and started matching it. The user had to say that
they did not want that site, they wanted consistent branding.
1. **Find out what the user likes about it.** Use their words. If they did not
say, ask once, or state your guess and carry on.
2. **Open it and look.** Load the page in a browser and capture it at a phone
width and a desktop width. A fetched summary or the HTML alone is not
looking. Check both color schemes if the site has them.
3. **Write down examples and mishaps.** Examples are what works, stated as a
principle with the measurement that makes it work. Mishaps are what fails.
Every site has both, including the ones we admire. Do not inherit a mishap
because it came with a look you like.
4. **Translate, do not copy.** Take the principle into this project's own
fonts, palette and components. Never lift a palette, typeface, layout or
copy.
5. **Report it.** Tell the user the two or three things you are taking and
what you are leaving behind.
Date every study. Sites change. Re-open the page before you rely on a
measurement recorded here.
## Sites and portfolios
| Reference | Study for | Start here |
|---|---|---|
| Joseph Zhang (leads design at Cognition; earlier Notion, Cursor, Skiff, Apple) | Restraint: a tiny type system, one accent, real work shown first | https://joseph.cv, https://joseph.cv/cursor, https://joseph.cv/reading |
### Joseph Zhang, joseph.cv
Checked 2026-11-21 at 1440 and 390 wide, dark and light preference. The home
page and one case study (`/cursor`) were opened. Measurements are computed
styles from the live page.
**Examples**
- **No hero.** The first screen holds name, role, city, one paragraph, the
teams he worked on and his links. Work starts directly under it. There is no
call-to-action button. "Get in touch" is a link inside the paragraph.
- **Two typefaces, three sizes.** A sans at 15px (14px on phones), weight 500,
for everything you read. A mono at 11.5px, uppercase and letterspaced, for
labels. 10px for small marks. Hierarchy comes from case and color, not from
bigger headings.
- **One accent, one meaning.** Yellow marks only two things: the arrow that
says a link leaves the site, and footnote numbers. Nothing else is yellow.
- **A plain grid.** Twelve columns, 12px gaps, 12px page margin. Work sits in
three columns of uneven-height tiles, and becomes one column on a phone with
no sideways scroll.
- **The real thing in every tile.** Each tile shows the actual product: a
screenshot or a short video on its own background. The caption is a title
and one category label. No description, tags, dates or buttons.
- **A case study in one screen.** One large image, two short paragraphs, then
facts as label and value pairs (role, collaborators, duration) and two plain
link rows. Collaborators are credited by name.
- **A version in the footer.** The site shows a version number and a
last-updated date, like a product that is maintained.
**Mishaps**
- **Labels are too faint.** The gray category labels are 11.5px at 3.91:1
against the page. The footer date is 2.20:1. Both are under 4.5:1.
- **Dark only.** A light color-scheme preference gets the same dark page.
- **Weak document structure.** Every project title is an `h1` (14 on the home
page). The `html` element has no `lang`. Every page has the same `<title>`.
Every image has the same alt text, "project hero".
- **Tight targets.** The stacked Teams and Links rows are 20px apart on a
phone, under the 24px minimum target size.
- **Scripts carry the media.** The raw HTML says "Loading..." in every tile
until scripts run. Four videos autoplay on the home page.
**Take:** cut type sizes before you add one. Give the accent color a single
meaning. Show the real product and let a title and one label caption it.
**Leave:** the faint labels, the dark-only theme and the heading structure.
Keep the restraint and meet [the shared standard](web-design.md) for contrast,
structure and targets.
## AI workspaces and agent products
The principles drawn from these are in
[the AI workspace design reference](ai-workspace-design.md). The links were
checked on 2026-10-26.
| Reference | Study for | Start here |
|---|---|---|
| Harvey II | Project workspace, reviewable agent output, evidence | https://www.harvey.ai/blog/introducing-harvey-ii, https://www.harvey.ai/blog/how-we-approach-design-at-harvey |
| Codex app (Ed Bayes, Andrew Ambrosino) | Supervising parallel agents, approvals, review queue | https://openai.com/index/introducing-the-codex-app/, https://softwareengineeringdaily.com/wp-content/uploads/2026/01/SED1898-OpenAI.txt |
| Linear (Karri Saarinen) | Work navigation, hierarchy, AI workbench model | https://linear.app/now/design-for-the-ai-age, https://karrisaarinen.com/posts/every-feature-should-earn-its-place/ |
| Figma (Rasmus Andersson) | Role- and selection-sensitive UI, pattern reduction | https://rsms.me/work/figma/, https://www.figma.com/blog/ui-refresh/ |
| Raycast | Fast, discoverable actions | https://www.raycast.com/blog/a-fresh-look-and-feel, https://www.raycast.com/why |
| Dust (Édouard Wautier, Clément Aupiais) | Human/agent symmetry, prototype-first, trustworthy data display | https://dust.tt/blog/field-study-prototypes-over-mockups, https://medium.com/algolia-design/building-trust-with-data-visualisation-13321b44b819 |
| Anthropic (Nate Parrott, Meaghan Choi, Jenny Wen, Joel Lewenstein) | Chat plus specialized UI, mental models, deciding what to build | https://www.aidesignfieldguide.com/articles/nate-parrott, https://designerfund.com/blog/ai-design-anthropic, https://www.linkedin.com/pulse/whats-worth-making-joel-lewenstein-xcgqc/ |
| Geist (Anthropic, Harvey brand) | Values expressed through type, palette and components | https://geist.co/work/anthropic |
## Adding a reference
Add a row to the table that fits. For a site you studied, add a dated section
with examples and mishaps like the one above. Keep only what you saw on the
page. Do not credit a whole product to one designer; public credits cover
specific work.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| brainstormingobra/superpowersYou MUST use this before any creative work - creating features, building components, adding functionality, modifying behavior, or planning anything new, in software or out of it (a talk, a business, a renovation). | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| brand-guidelinesanthropics/skillsApplies Anthropic's official brand colors and typography to any sort of artifact that may benefit from having Anthropic's look-and-feel. Use it when brand colors or style guidelines, visual formatting, or company design standards apply. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| frontend-designanthropics/skillsGuidance for distinctive, intentional visual design when building new UI or reshaping an existing one. Helps with aesthetic direction, typography, and making choices that don't read as templated defaults. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| mcp-builderanthropics/skillsGuide for creating high-quality MCP (Model Context Protocol) servers that enable LLMs to interact with external services through well-designed tools. Use when building MCP servers to integrate external APIs or services, whether in Python (FastMCP) or Node/TypeScript (MCP SDK). | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |