본문으로 건너뛰기
cccskills
무료GitHub에 공개

design-archetype

Pick a product archetype from a curated catalog before any UI gets built — the single biggest lever for not looking vibecoded. Routes the request to one of B2B-productivity, enterprise-trust, premium-consumer, playful-consumer, editorial, brutalist-distinctive, dev-tool, or marketing-landing, and returns a complete design philosophy: typography pair, color logic, motion curve, density, icon stance, reference sites, and a "feels like X" claim. Load when the user asks to pick an aesthetic, choose a design direction, decide what a UI should feel like, classify a product visually, says "what should this look like", "make this feel like [Linear/Stripe/Apple/Duolingo/...]", "give me a design direction", "what's the right aesthetic for this", or when frontend-design routes here during archetype selection. Sub-skill of frontend-design.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

연동 도구NotionStripeVercel
설치 방법 보기

포함된 파일(15)

  • SKILL.md8.3 KB
  • DEPRECATION.md994 B
  • references/archetypes/b2b-productivity.md3.2 KB
  • references/archetypes/brutalist-distinctive.md2.8 KB
  • references/archetypes/conversational-ai.md3.9 KB
  • references/archetypes/creative-tool.md3.5 KB
  • references/archetypes/dev-tool.md2.8 KB
  • references/archetypes/editorial.md2.8 KB
  • references/archetypes/enterprise-trust.md3.2 KB
  • references/archetypes/marketing-landing.md3.6 KB
  • references/archetypes/playful-consumer.md2.9 KB
  • references/archetypes/premium-consumer.md3.1 KB
  • references/archetypes/social-feed.md3.6 KB
  • references/archetypes/spatial-canvas.md3.7 KB
  • references/selection-rubric.md4.1 KB

SKILL.md(원문)

설치하기 전에 에이전트에게 전달되는 지시 내용을 확인할 수 있습니다.

Design Archetype

You are the Design Direction Lead. You take a product description and pick exactly one archetype from a curated catalog. The archetype is not a vibe — it is a complete design philosophy that downstream skills (design-tokens-craft, icon-craft, frontend-design) consume to generate non-generic output.

Hard Rules

  • Pick exactly one archetype. Hybrids look vibecoded. If two fit, pick the dominant one and note the secondary as influence only.
  • Always ground with a reference. Every archetype output must include a feels like X claim naming a real, current product.
  • Never invent archetypes. If nothing in the catalog fits, the closest match wins and gets adapted — do NOT invent a new philosophy mid-skill.
  • No placeholder reference sites. Reference sites must be real, current, and accessible — verify before naming them.

Workflow

Step 1 — Read the Request

Extract: product type, audience, emotional goal, any reference products the user named, technical constraints. If the user named a reference product, that almost always determines the archetype directly.

Step 2 — Score Each Archetype 0–3

Read references/selection-rubric.md. Score the 12 archetypes against the request on three dimensions:

  • Audience fit — does this audience expect this aesthetic?
  • Job fit — does this aesthetic serve the product's primary job?
  • Distinctive fit — does this aesthetic give the product an edge over generic competitors?

Highest combined score wins. Ties → pick the more specific (less generic) archetype.

Step 3 — Load the Archetype File

Open references/archetypes/<archetype>.md. It contains:

  • Typography pair (display + body, with weights and sizes)
  • Color logic (background story, accent strategy, dark mode story)
  • Motion philosophy (duration, easing, character)
  • Density (information per screen)
  • Icon stance (handoff to icon-craft)
  • Layout signatures (the moves that recur)
  • 3 reference sites
  • Anti-patterns (what NOT to do in this archetype)

Step 4 — Customize for the Specific Product

Adapt the archetype using the user's specifics:

  • If audience is highly technical → tighten density, reduce ornamentation
  • If product is regulated (finance, health) → shift palette toward muted, raise contrast
  • If product is creative-tools → loosen motion budget, add character
  • If product is mobile-first consumer → enlarge tap targets, re-tune type scale

Record adaptations as 2–4 bullets, not a rewrite.

Step 5 — Write ARCHETYPE.md

Write to .design/<feature>/ARCHETYPE.md using the Output Format below.

Step 6 — Hand Off

Return the archetype name + the path to ARCHETYPE.md. frontend-design reads it for Step 4 (design-tokens-craft) and Step 5 (icon-craft).


The Catalog (12 archetypes)

ArchetypeUse whenFeels like
b2b-productivityInternal tools, dashboards, dense workflow apps for power usersLinear, Stripe Dashboard, Vercel
enterprise-trustRegulated, finance, healthcare, ops where credibility > delightNotion (formal), Atlassian, Datadog, Ramp
premium-consumerConsumer products charging premium, focused on craft and calmApple, Arc, Things 3, Bear, Linear marketing
playful-consumerConsumer products optimizing for delight, retention through emotionDuolingo, early Notion, Figma, Tumblr 2014, Calm
editorialContent-heavy, reading-first, story-led productsSubstack, Medium 2014, NYT, Stripe Press, Are.na
brutalist-distinctiveProducts that win by being unapologetically themselvesGumroad, Are.na, Vercel labs, Posthog, Cron 2022
dev-toolCLIs, terminals, dev infrastructure, IDE-adjacentWarp, Raycast, Fly.io, Railway, Anthropic Console
marketing-landingOne-shot marketing site for an existing productVercel, Resend, Linear marketing, Stripe marketing
creative-toolAI media studios, generative art tools, prosumer creative appsLeonardo.ai, Midjourney, Runway, Krea, Suno, Luma
social-feedNetworked-attention products, feed-first consumer/prosumerX, Threads, Bluesky, Reddit, Mastodon, BeReal
conversational-aiChat-first AI products where prompt + response is the canvasChatGPT, Claude, Perplexity, Gemini, Poe
spatial-canvasInfinite-canvas thinking/diagramming/workshop toolsFigJam, Miro, tldraw, Whimsical, Excalidraw

Output Format (ARCHETYPE.md)

# Archetype: [name]

**Feels like:** [reference product]
**Why this archetype:** [2 sentences max]

## Typography
- Display: [font, weight, sizes]
- Body: [font, weight, sizes]
- Mono (if applicable): [font, weight, sizes]
- Pairing rationale: [1 sentence]

## Color
- Background story: [light + dark]
- Foreground story: [light + dark]
- Accent strategy: [primary, secondary, when each appears]
- Anti-defaults: [explicitly forbidden colors for this build]

## Motion
- Duration budget: [N–N ms]
- Easing curve: [cubic-bezier or named]
- Character: [restrained / characterful / theatrical]

## Density
[dense / standard / generous]

## Icon stance
[custom-svg / tuned-phosphor / hand-drawn / mixed-metaphor / system-native]

## Layout signatures
- [move 1]
- [move 2]
- [move 3]

## Reference sites (visit these)
1. [URL] — [what to study]
2. [URL] — [what to study]
3. [URL] — [what to study]

## Adaptations for this product
- [bullet]
- [bullet]

## Anti-patterns (do NOT do these even if instinct says to)
- [bullet]
- [bullet]

Gotchas

  • "B2B SaaS" is NOT an archetype — it's a market. A B2B SaaS for ops engineers is dev-tool; a B2B SaaS for finance leaders is enterprise-trust; a B2B SaaS for designers is premium-consumer. Always go one layer deeper.
  • The user saying "make it look modern" is meaningless. Pick the archetype anyway based on audience and job, then defend the choice.
  • "Make it playful but professional" almost always means b2b-productivity with a slightly warmer palette, NOT playful-consumer. The latter is for products where delight is the actual job.
  • Brutalist is overused as a fallback when designers don't want to commit. Use it only when the brand identity ALREADY has a brutalist position — it cannot be retrofitted.

Reference Files

  • references/selection-rubric.md — scoring rubric, tiebreakers, decision tree, common-confusion guide
  • references/archetypes/<name>.md — one file per archetype in the catalog above. Open the file matching your selection in Step 3. Each contains: typography pair, color logic, motion philosophy, density, icon stance, layout signatures, 3 reference sites, anti-patterns.

Impact Report

Archetype selected: [name]
Feels like: [reference product]
Top scores: [archetype: N/9, archetype: N/9, archetype: N/9]
Adaptations applied: [count]
ARCHETYPE.md written to: .design/<feature>/ARCHETYPE.md
Handoff to: design-tokens-craft (Step 4 of frontend-design)

리뷰

아직 리뷰가 없습니다. 사용해 본 소감을 남겨 주세요.

댓글

로그인 없이 쓸 수 있습니다. 사용 후기, 질문, 작동하지 않은 환경 등을 편하게 남겨 주세요.

아직 댓글이 없습니다. 첫 댓글을 남겨 보세요.

보내면 바로 공개됩니다. 이메일 주소나 전화번호 같은 개인 정보는 쓰지 마세요. URL은 2개까지입니다.

0 / 1,000자

이 양식은 reCAPTCHA로 보호되며 Google 개인정보처리방침 및 서비스 약관이 적용됩니다.

같은 저장소의 스킬

Put on the adversarial hat and systematically attack any document, plan, strategy, or idea to expose its weakest points before commitment. Structured devil's advocate with red team rigour — not pessimism, but evidence-based critique across three phases: diagnostic (are claims accurate?), creative (is the problem artificially constrained?), challenge (are solutions robust?). Load when the user asks to stress test a document, red team this plan, poke holes in this, devil's advocate this, challenge my assumptions, or when product-soul, brainstorming, prd-writing, or inversion calls for adversarial review. Also triggers on "what am I missing", "what could kill this", "find the flaws", or "critique this rigorously".

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

dvy1987/agent-loom32026년 8월 8일 업데이트

Design execution structure for decomposed processes: single agent or multi-agent topology. Load when user says "design an agent for this", "what agent structure do I need", "architect this", "should this be multi-agent", "what's the right execution structure", "agent topology", "how should agents be organized". Takes process-decomposer output as primary input. If triggered directly without a process entry, calls process-decomposer first.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

dvy1987/agent-loom32026년 8월 8일 업데이트

Internal skill. Called by setup-evaluation after a PASS. Launches agents from a validated architecture spec using Claude Code / Ampcode native parallelism (Task tool). Does NOT generate scripts or SDK code — it outputs structured spawn instructions that the platform executes natively. Never invoked directly by the user. Never launches without a setup-evaluation PASS.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

dvy1987/agent-loom32026년 8월 8일 업데이트

Sync library skills from an agent-loom upstream repo into this project's .agents/skills while preserving project-local and forked skills. Load when the user asks to sync agent-loom, update skills from upstream, rsync from ../agent-loom, pull new library skills, upgrade installed skills, or refresh the .agents folder without losing custom project skills. Also triggers on "sync skills from agent-loom", "update my agent skills", "pull skill library updates", or "merge agent-loom improvements into this repo".

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

dvy1987/agent-loom32026년 8월 8일 업데이트

Instrument a shipped product's AI agents with tracing and observability so you can see what they did, why outputs happened, and what each run cost. Plain-language primer plus free-tier-first backend selection (Langfuse, Phoenix, LangSmith, Braintrust) and OpenTelemetry/OpenInference instrumentation. Load when the user asks to add observability, add tracing, instrument my agents, see what my agent is doing in production, set up Langfuse or Phoenix or LangSmith, debug why my agent gave a bad answer, or track LLM cost per request. Also fires when agent-system-architecture or setup-evaluation requires an observability plan for an agent-chain product. NOT for tracing the coding agent itself — that is run-trace. Precondition for runtime-learning-loop.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

dvy1987/agent-loom32026년 8월 8일 업데이트

Run a structured retrospective after development-phase runs of your product's agents — interview the owner in plain language about what went well and poorly, draft ranked improvement hypotheses, then design and run small n=1/n=2 experiments with pre-declared success criteria, guardrails, stop conditions, and a cost/ROI kill-switch. Load when the user says how did that run go, retro this run, the agent output was bad, what should we improve, draft hypotheses, run a small experiment, or after repeated dev runs of an agentic system produce uneven quality. Priority: output quality over performance over cost, each with diminishing-returns stops. NOT a product A/B test (experimentation), NOT coding-agent harness repair (harness-evolution), NOT production-scale learning (runtime-learning-loop).

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

dvy1987/agent-loom32026년 8월 8일 업데이트

dvy1987의 스킬 모두 보기

이 스킬의 문제 신고