use-design-md
Skill78 starsChanged 6 days agoMITView on GitHub
Apply a Korean brand's published DESIGN.md from the ko-design-md catalog (getdesign.kr) — colors, typography, spacing, radius, components, do's & don'ts — to the UI you are building in the CURRENT project. Use this skill whenever the user wants UI built or restyled in the *style of* a catalogued Korean service — "토스 디자인으로 만들어줘", "당근 스타일로 이 화면 다시 꾸며줘", "getdesign 카탈로그에서 배민 디자인 가져와서 적용", "KRDS 톤으로 폼 잡아줘", "make this look like Toss". Works in any repository over the network. Do NOT use it to add or edit catalog entries (that is the separate `design-md` producer skill), to explain the DESIGN.md format itself — there is no UI to restyle there — or to build a quiz or exam page that only borrows a brand's tone.
In this repository5 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- design-mdSkill
- docs-crawlerSkill
- preview-prose-auditSkill
The file
We split this skill into the rules it gives an agent
7 rules. About 0 are common, 2 are distinctive, and 5 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
5
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 16
Every catalog fetch (index, entry, token sidecar, GitHub raw fallback) follows these rules.
- Very specific: line 33
Read the file to the end: Check its size (wc -lc <file>), then read it with your file-reading tool.
- Very specific: line 50
If this fails you have no index — that is not a No match.
- Distinctive: line 86
Read it to its last line before applying anything — the prose carries intent (do's & don'ts, known gaps) that the tokens alone don't.
- Very specific: line 93
Before editing anything, read references/apply-guide.md and follow it.
- Distinctive: line 99
Don't gate on the current repo — this skill runs anywhere.
- Very specific: line 108
Stay vendor-neutral: keep the source's name — brand or design system — out of the generated UI (references/apply-guide.md §6).
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
use-design-md — consumer skill for the ko/design.md catalog
The ko-design-md catalog (https://www.getdesign.kr) publishes one DESIGN.md per Korean
brand: colors in OKLCH, typography, spacing, radius, components, and do's & don'ts. This
skill pulls the right entry and uses it as the design brief for UI work in the current
project. Invoked by name it is /use-design-md (skills.sh) or
/ko-design-md:use-design-md (Claude Code plugin marketplace).
Fetching — files, failures, the whole document
Every catalog fetch (index, entry, token sidecar, GitHub raw fallback) follows these
rules. Fetch with curl, which writes the response body to a file byte for byte. A
web-reading tool that passes the page through a model (Claude Code: WebFetch)
summarizes and drops exact values (an OKLCH triple, a 13px step); use one only when you
have no shell at all.
- Save to a file, don't print to the terminal. Entries run to tens of kilobytes, and
a shell tool caps the output it hands back (Claude Code: around 30,000 characters),
silently dropping the back half — components, do's & don'ts, known gaps. Download into
your per-user cache,
"${XDG_CACHE_HOME:-$HOME/.cache}/use-design-md/"(not the user's project, not a shared/tmp);--create-dirsmakes it, so each fetch stays one plaincurlcommand (a compoundmkdir …; curl …is refused where onlycurlis allowed to run). - Make HTTP failures fail.
curl -fsSL --create-dirs -o <file> <url>.-fturns a 404/5xx into a non-zero exit — without it a missing slug's one-lineNot found: <slug>body arrives as if it were the document. After a non-zero exit you did not get the file; what's on disk may be stale or partial — don't read it. - Read the file to the end. Check its size (
wc -lc <file>), then read it with your file-reading tool. That tool may cap a single read too, so when one read doesn't reach the end, read the rest in chunks (Claude Code:offset/limit). You're done when you've seen the last line numberwc -lreported. If your file reader expects a different path form than your shell prints, convert it (Git Bash on Windows:cygpath -w <file>).
The references/ files this skill points to sit beside it. If you have only this file
(e.g. loaded from getdesign.kr's /.well-known/agent-skills/), fetch them the same way
from https://raw.githubusercontent.com/CaesiumY/ko-design-md/main/.claude/skills/use-design-md/references/<name>.md.
Step 1 — Discover: resolve the brand to a slug
curl -fsSL --create-dirs -o "${XDG_CACHE_HOME:-$HOME/.cache}/use-design-md/index.txt" https://www.getdesign.kr/llms.txtIf this fails you have no index — that is not a No match. Use the index fallbacks in
references/endpoints.md; if those fail too, tell the user you couldn't reach the
catalog — don't report the brand as missing.
Each entry line looks like:
- [당근](https://www.getdesign.kr/services/seed-design/llms.txt): community · SEED Design — <tagline>An entry that publishes a design-system name carries it after the category, as above;
one that doesn't has the category alone (: finance — <tagline>). Match the user's
mention against the link text (name), the design-system name, AND the slug in the URL.
Most names are Korean and some slugs are design-system names (당근 → seed-design,
구름 → vapor-ui), so translate an English brand name to the Korean name first
(Karrot → 당근).
- One clear match → Step 2.
- Several plausible matches → ask the user which one they mean (Claude Code:
AskUserQuestion). - No match → tell the user plainly the brand isn't in the catalog, optionally list a
few catalogued brands in the nearest category, and note that adding it is the
design-mdskill's job (see Scope guardrails).
Step 2 — Fetch the DESIGN.md
curl -fsSL --create-dirs -o "${XDG_CACHE_HOME:-$HOME/.cache}/use-design-md/<slug>.md" https://www.getdesign.kr/services/<slug>/llms.txt- 404 → the slug isn't in the catalog: take the No match path rather than guessing another slug.
- Any other failure (network, 5xx, 403/429) says nothing about the catalog: try the
GitHub raw fallback in
references/endpoints.md; if that fails too, tell the user you couldn't fetch the entry — don't report the brand as missing.
Read it to its last line before applying anything — the prose carries intent (do's &
don'ts, known gaps) that the tokens alone don't. If you need tokens as structured data
(to generate a theme programmatically), fetch the JSON sidecar the same way — it is on
GitHub raw, not getdesign.kr (references/endpoints.md §3).
Step 3 — Apply to the current project
Before editing anything, read references/apply-guide.md and follow it. Three parts are
never optional: the entry's Known Gaps come first (apply-guide §2), its Do's and Don'ts are
hard constraints, and you verify the result (apply-guide §5) before calling it done.
Scope guardrails
- Don't gate on the current repo — this skill runs anywhere.
- Adding or editing a catalog entry is the
design-mdproducer skill's job, inside the ko-design-md repo. Stop and point the user there. - Don't invent the design. A brand that isn't listed isn't available here — say so instead of writing a DESIGN.md for it or approximating it from memory. For a listed brand, don't invent values its DESIGN.md doesn't have: propose an extension marked as your inference, not the brand's spec.
- Loaded for a page that only wants a brand's tone ("토스 앱처럼 깔끔한 퀴즈")? Still fetch the entry rather than approximating the brand from memory.
- Stay vendor-neutral: keep the source's name — brand or design system — out of the
generated UI (
references/apply-guide.md§6).
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

