Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →OmD Lab #01 — DESIGN.md 유무·생성방식이 UI 결과물에 미치는 영향을 4가지 조건으로 비교. 토스 스타일 모바일 UI를 (V1) DESIGN.md 없이, (V2) 수동 작성, (V3) 자동 생성, (V4) 자동 생성 + 5회 피드백 루프로 만들어 동시 비교 뷰로 시각화. 사용자가 본인의 Claude Code 안에서 재현 가능한 실험.
.claude/skills/kwakseongjae-omd-lab-01-designmd-impact/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | -68% | 0% |
| case-02 | ✗→✓ | ▲ Improved | -25% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 127% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -65% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -15% | 0% |
CORE_V2_LEGACY_EXPERIMENT_BLOCKED
The historical Lab #01 playbooks authored retired 9/12-section DESIGN.md files. They remain evidence about an earlier format, not a valid template for new runs. Do not create, update, or normalize a DESIGN.md through those playbooks.
For a new format-impact experiment, preregister arms against DESIGN.md Core v2: model-only, standalone Portable Core, and an adopted Bound System where relevant. Use the same sealed prompt/model/effort/budget and keep full-harness comparisons separate. A new visible document must have a clean top and the exact seven Core anchors, keep unknowns absent, and be useful without OmD or sidecars. Only an adopted profile: portable-core manifest may make its graph canonical.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 22,160 | 6,480 | -71% | 1 | 1 | 0% | 3,731 | 1,181 | -68% | 0 | 0 | — |
case-02 | fail→pass | 17,479 | 10,803 | -38% | 1 | 1 | 0% | 2,888 | 2,155 | -25% | 0 | 0 | — |
case-03 | fail→pass | 3,575 | 6,287 | +76% | 1 | 1 | 0% | 562 | 1,274 | +127% | 0 | 0 | — |
case-04 | fail→pass | 20,518 | 6,585 | -68% | 1 | 1 | 0% | 3,940 | 1,385 | -65% | 0 | 0 | — |
case-05 | fail→pass | 10,238 | 6,487 | -37% | 1 | 1 | 0% | 1,695 | 1,434 | -15% | 0 | 0 | — |
case-06 | fail→pass | 18,199 | 10,596 | -42% | 1 | 1 | 0% | 2,934 | 2,007 | -32% | 0 | 0 | — |
case-07 | fail→pass | 10,764 | 3,988 | -63% | 1 | 1 | 0% | 1,809 | 922 | -49% | 0 | 0 | — |
case-08 | fail→pass | 16,636 | 10,056 | -40% | 1 | 1 | 0% | 2,798 | 2,041 | -27% | 0 | 0 | — |
case-09 | fail→pass | 13,848 | 5,213 | -62% | 1 | 1 | 0% | 2,188 | 936 | -57% | 0 | 0 | — |
case-10 | fail→pass | 31,520 | 1,713 | -95% | 1 | 1 | 0% | 1,629 | 435 | -73% | 0 | 0 | — |
case-11 | fail→pass | 5,226 | 1,866 | -64% | 1 | 1 | 0% | 826 | 507 | -39% | 0 | 0 | — |
case-12 | fail→pass | 10,473 | 3,412 | -67% | 1 | 1 | 0% | 1,743 | 831 | -52% | 0 | 0 | — |
case-13 | fail→pass | 49,148 | 1,903 | -96% | 1 | 1 | 0% | 5,169 | 522 | -90% | 0 | 0 | — |
case-14 | pass→pass | 13,029 | 3,450 | -74% | 1 | 1 | 0% | 2,286 | 844 | -63% | 0 | 0 | — |
case-15 | fail→pass | 15,943 | 7,171 | -55% | 1 | 1 | 0% | 2,735 | 1,454 | -47% | 0 | 0 | — |
case-16 | pass→pass | 14,629 | 4,891 | -67% | 1 | 1 | 0% | 2,411 | 1,001 | -58% | 0 | 0 | — |
case-17 | pass→pass | 17,088 | 1,840 | -89% | 1 | 1 | 0% | 3,009 | 500 | -83% | 0 | 0 | — |
case-18 | fail→pass | 13,125 | 4,806 | -63% | 1 | 1 | 0% | 2,307 | 994 | -57% | 0 | 0 | — |
case-19 | pass→pass | 9,251 | 3,078 | -67% | 1 | 1 | 0% | 1,453 | 736 | -49% | 0 | 0 | — |
case-20 | pass→pass | 12,686 | 9,124 | -28% | 1 | 1 | 0% | 2,491 | 1,919 | -23% | 0 | 0 | — |
case-21 | pass→pass | 7,615 | 4,756 | -38% | 1 | 1 | 0% | 1,406 | 1,045 | -26% | 0 | 0 | — |
case-22 | pass→pass | 10,442 | 7,599 | -27% | 1 | 1 | 0% | 1,794 | 1,458 | -19% | 0 | 0 | — |
case-23 | pass→pass | 8,163 | 6,886 | -16% | 1 | 1 | 0% | 1,568 | 1,499 | -4% | 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 +65 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.
| Model | Method | Date | Lift |
|---|---|---|---|
| gemini-3.6-flash | verified | 8/20/2026 | +64% |
Other measured skills in the registry, with their headline benchmark lift.