Install any skill in seconds. Free to start, no credit card required.
Get Started Free →/design Path A 워크플로의 Claude Design 핸드오프 패키지 전문가. 붙여넣기 바로 쓰는 claude.com Design 세션용 5파일 핸드오프 번들(prompt/context/references/acceptance/checklist)을 조립합니다. 브랜드 미비 폴백과 섹션 재생성을 처리합니다.
.claude/skills/modu-ai-design-handoff/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-09 | ✗→✓ | ▲ Improved | 119% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 197% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 31% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 463% | 0% |
| case-11 | ✗→✓ | ▲ Improved | 279% | 0% |
> ⚠️ 개발 런타임 전용 — 이 스킬은 MoAI-ADK(Claude Code) 환경을 전제한다. Claude Cowork(Desktop)에서는 .moai/config 의존으로 동작하지 않을 수 있다. Desktop 사용자는 moai-designer:design-handoff-reader를 사용한다.
<!-- Verifies: prompt.md is paste-ready (no MoAI tokens) --> <!-- Verifies: Brand voice integrated when present; graceful default when absent --> <!-- Verifies: Handoff exit AskUserQuestion with 3 options -->
<!-- @MX:ANCHOR: AUTO] 5-section prompt.md template structure — canonical definition --> <!-- @MX:REASON: Consumed by every /design Path A handoff execution (high fan_in). Structural changes affect user trust — prompt.md is pasted directly into external claude.com Design session. -->
Assembles the 5-file Claude Design handoff package for the /design Path A workflow (the handoff-deliverable step). The package is designed for paste-and-go use in the external claude.com Design product.
The handoff package lives at .moai/design/handoff/:
| File | Purpose | Paste target | |------|---------|-------------| | prompt.md | Master prompt — paste directly into claude.com Design | Yes (primary) | | context.md | Extended context for reference during design session | Optional supplement | | references.md | Visual reference URLs and design inspiration sources | Referenced in prompt | | acceptance.md | Design acceptance criteria (WCAG, responsive, brand) | Referenced in prompt | | checklist.md | Pre-paste self-check before using in claude.com Design | Human review tool |
Key guarantees:
prompt.md contains NO MoAI-specific tokens (no SPEC-, .moai/, manager-, internal skill names).moai/project/brand/brand-voice.md existsBrand Voice (default — please customize) placeholder sectiondesign-handoff-reader (the preceding Path A step) — what the user wants designed, target users, value propositionsdesign-brief when the request originates from a design brief (Lean Canvas / evaluation context).moai/project/brand/brand-voice.md (brand context).moai/project/brand/visual-identity.md (design tokens, colors)Before writing any file, check brand context:
IF .moai/project/brand/brand-voice.md exists AND is non-empty:
Load brand voice → use in Brand Voice section of prompt.md
SET brand_present = true
ELSE:
Use default brand voice placeholder
SET brand_present = false
Note: will include AskUserQuestion offer to run brand interview<!-- @MX:WARN: AUTO] prompt.md template — output pasted into external claude.com Design session --> <!-- @MX:REASON: Changes to this template affect what users paste into claude.com. Structural changes can break user's design sessions. Validate against current claude.com Design prompt guidelines before modifying. -->
prompt.md MUST follow this exact 5-section structure:
brand_present = true): Extract personality + voice guidelines + color palette + typography from brand-voice.md and visual-identity.mdbrand_present = false): Emit ## 3. Brand Voice (default — please customize) header with explicit placeholder + instructions to either edit or run brand interviewSee 5-section prompt template + brand branches detail for verbatim section templates.
HARD] The following MUST NOT appear anywhere in prompt.md:
SPEC- identifiers (e.g., SPEC-AUTH-001).moai/ paths (e.g., .moai/design/, .moai/project/)manager-spec, design-brief, design-handoff-reader)/design, /moai plan)The prompt must read as if written by a human product designer with no knowledge of MoAI's internal structure.
| Step | File | Purpose | |------|------|---------| | 2 | references.md | Competitor analysis + visual inspiration + UX pattern references (3-5 URLs from the bundle's Sources). Falls back to instructional note when URLs are scarce. | | 3 | acceptance.md | Accessibility (WCAG 2.1 AA), Responsiveness (375/768/1280px), Brand Alignment, Content Completeness, Technical Constraints | | 4 | context.md | Extended context — NOT for pasting into Claude Design. Full brief summary, roadmap context, research findings, brand context | | 5 | checklist.md | Human self-check before pasting prompt.md: content review, MoAI-internal cleanup (auto-verified), scope verification, session readiness |
See supporting files templates for verbatim references.md, acceptance.md, context.md, and checklist.md templates.
After all 5 files are written, the skill MUST surface an AskUserQuestion (with ToolSearch preload) presenting 3 options:
ToolSearch(query: "select:AskUserQuestion")
AskUserQuestion({
questions: [{
question: "핸드오프 패키지가 준비되었습니다. 다음 단계를 선택하세요.",
header: "Design 핸드오프 완료",
options: [
{
label: "Claude Design 세션 시작 (권장)",
description: "prompt.md를 claude.com Design에 복사해 붙여넣고 디자인 세션을 시작합니다. context.md/references.md는 보조 자료로 함께 엽니다."
},
{
label: "수동 검토",
description: "핸드오프 파일을 직접 검토하고 필요한 경우 편집합니다. .moai/design/handoff/ 디렉토리를 확인하세요. 준비가 되면 prompt.md를 Claude Design에 붙여넣으세요."
},
{
label: "핸드오프 패키지 재생성",
description: "prompt.md 또는 다른 파일에 수정이 필요한 경우 어떤 부분을 변경할지 알려주세요. 해당 파일만 재생성합니다."
}
]
}]
})For non-Korean conversation_language, translate option labels and descriptions accordingly.
design-brief: Produces the design brief (Lean Canvas + value propositions) consumed as primary input when the request originates from a briefdesign-system-prep / design-system-library: Provides reference URLs and visual inspiration for the references.md Sourcesdesign-handoff-reader: Preceding Path A step — summarizes the imported bundle into the inputs this skill consumesdesign-workflow: Downstream consumer of the .moai/design/handoff/ directory after the user completes the external Claude Design session (Path A handler)| Rationalization | Reality | |----------------|---------| | "Including SPEC-AUTH-001 in prompt.md helps the designer understand scope" | prompt.md is for claude.com Design, not MoAI. SPEC IDs are internal. Use the Out of Scope section to describe scope boundaries in plain English. | | "I should skip checklist.md — it's obvious" | Checklist.md prevents the most common error: pasting a prompt with placeholder Brand Voice. It takes 30 seconds to complete and saves a bad design session. | | "references.md is optional if research had no URLs" | references.md is always produced. When URLs are scarce, include a note asking the user to add their own. An empty references file is worse than one with instructions. | | "If brand is absent, skip the Brand Voice section" | Brand Voice section is always present. Brand-absent path produces an explicit placeholder with instructions — clearer than a missing section. |
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-09 | fail→pass | 8,674 | 10,291 | +19% | 1 | 1 | 0% | 1,482 | 3,249 | +119% | 0 | 0 | — |
case-07 | fail→pass | 15,273 | 22,664 | +48% | 1 | 1 | 0% | 1,884 | 5,591 | +197% | 0 | 0 | — |
case-01 | fail→fail | 35,924 | 17,428 | -51% | 1 | 1 | 0% | 5,915 | 2,798 | -53% | 0 | 0 | — |
case-08 | fail→fail | 11,071 | 14,058 | +27% | 1 | 1 | 0% | 1,711 | 3,805 | +122% | 0 | 0 | — |
case-02 | fail→pass | 30,309 | 34,797 | +15% | 1 | 1 | 0% | 4,597 | 6,002 | +31% | 0 | 0 | — |
case-03 | fail→fail | 22,878 | 12,280 | -46% | 1 | 1 | 0% | 4,034 | 2,612 | -35% | 0 | 0 | — |
case-04 | fail→fail | 12,943 | 17,969 | +39% | 1 | 1 | 0% | 1,147 | 4,177 | +264% | 0 | 0 | — |
case-05 | pass→pass | 12,539 | 17,735 | +41% | 1 | 1 | 0% | 1,134 | 4,459 | +293% | 0 | 0 | — |
case-06 | pass→pass | 29,699 | 22,673 | -24% | 1 | 1 | 0% | 5,588 | 6,434 | +15% | 0 | 0 | — |
case-10 | fail→pass | 13,729 | 34,587 | +152% | 1 | 1 | 0% | 1,490 | 8,393 | +463% | 0 | 0 | — |
case-11 | fail→pass | 8,118 | 19,137 | +136% | 1 | 1 | 0% | 1,278 | 4,840 | +279% | 0 | 0 | — |
case-12 | fail→pass | 12,312 | 9,673 | -21% | 1 | 1 | 0% | 2,314 | 3,074 | +33% | 0 | 0 | — |
case-13 | fail→pass | 14,877 | 14,564 | -2% | 1 | 1 | 0% | 2,354 | 3,841 | +63% | 0 | 0 | — |
case-14 | fail→pass | 20,396 | 22,361 | +10% | 1 | 1 | 0% | 2,750 | 5,055 | +84% | 0 | 0 | — |
case-15 | fail→pass | 9,450 | 8,477 | -10% | 1 | 1 | 0% | 1,441 | 2,854 | +98% | 0 | 0 | — |
case-16 | fail→fail | 13,227 | 8,822 | -33% | 1 | 1 | 0% | 1,467 | 2,934 | +100% | 0 | 0 | — |
case-17 | fail→pass | 15,546 | 13,035 | -16% | 1 | 1 | 0% | 1,902 | 3,132 | +65% | 0 | 0 | — |
case-18 | fail→fail | 14,767 | 8,649 | -41% | 1 | 1 | 0% | 2,897 | 3,002 | +4% | 0 | 0 | — |
case-19 | fail→pass | 19,311 | 11,460 | -41% | 1 | 1 | 0% | 2,279 | 3,323 | +46% | 0 | 0 | — |
case-20 | pass→pass | 10,211 | 6,849 | -33% | 1 | 1 | 0% | 1,662 | 3,390 | +104% | 0 | 0 | — |
case-21 | fail→fail | 11,575 | 7,055 | -39% | 1 | 1 | 0% | 890 | 2,563 | +188% | 0 | 0 | — |
case-22 | pass→pass | 12,652 | 8,539 | -33% | 1 | 1 | 0% | 2,210 | 3,010 | +36% | 0 | 0 | — |
case-23 | fail→pass | 13,865 | 9,909 | -29% | 1 | 1 | 0% | 1,418 | 3,229 | +128% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 23 cases were attempted, and 21 counted toward the lift figure. The other 2 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +52 percentage points is the difference between those two pass rates over the 21 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.