Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →URL 또는 brand id 입력 → 3-tier 검증 파이프라인(Tier 1 라이브 + Tier 2 getdesign/refero + Tier 3 reconcile)으로 references/<id>/DESIGN.md를 신규 생성(CREATE)하거나 기존 섹션을 검증·갱신(UPDATE). '레퍼런스 추가/수정', 'X DS 검증', 'X 컴포넌트 다시 뽑아줘' 류에 트리거.
.claude/skills/kwakseongjae-omd-add-reference/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-05 | ✗→✓ | ▲ Improved | 40% | 0% |
| case-01 | ✗→✓ | ▲ Improved | -9% | 0% |
| case-02 | ✗→✓ | ▲ Improved | -6% | 0% |
| case-03 | ✗→✓ | ▲ Improved | -22% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -61% | 0% |
Status: CORE_V2_CATALOG_WRITE_BLOCKED.
The historical reference catalog remains a dual-read migration input. It is not permission to create another legacy file. Do not write or update web/references/<id>/DESIGN.md, frontmatter tokens, generated counts, or mirrors.
On invocation, return the blocker and the proposed reference id/source list without writing. Authoring may resume only after a graph-backed reference metadata/evidence schema, deterministic Core projection, migrated registry and quality readers, legacy dual-read/Core single-write tests, and a dropped-zero migration gate all pass together.
Never restore a legacy YAML marker, 13/15/16-section output, placeholder sentinels, or a second authoritative Markdown dialect as a shortcut.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-05 | fail→pass | 3,791 | 3,416 | -10% | 1 | 1 | 0% | 624 | 871 | +40% | 0 | 0 | — |
case-22 | pass→pass | 21,640 | 12,963 | -40% | 1 | 1 | 0% | 2,448 | 2,716 | +11% | 0 | 0 | — |
case-23 | pass→pass | 23,174 | 21,479 | -7% | 1 | 1 | 0% | 4,122 | 4,755 | +15% | 0 | 0 | — |
case-01 | fail→pass | 68,610 | 32,663 | -52% | 1 | 1 | 0% | 881 | 805 | -9% | 0 | 0 | — |
case-02 | fail→pass | 6,018 | 3,896 | -35% | 1 | 1 | 0% | 1,078 | 1,018 | -6% | 0 | 0 | — |
case-03 | fail→pass | 8,765 | 2,966 | -66% | 1 | 1 | 0% | 1,024 | 800 | -22% | 0 | 0 | — |
case-04 | fail→pass | 10,428 | 2,785 | -73% | 1 | 1 | 0% | 1,947 | 769 | -61% | 0 | 0 | — |
case-06 | fail→pass | 27,407 | 3,053 | -89% | 1 | 1 | 0% | 5,034 | 785 | -84% | 0 | 0 | — |
case-07 | fail→pass | 7,100 | 3,632 | -49% | 1 | 1 | 0% | 1,203 | 935 | -22% | 0 | 0 | — |
case-08 | fail→pass | 9,255 | 3,773 | -59% | 1 | 1 | 0% | 1,926 | 789 | -59% | 0 | 0 | — |
case-09 | fail→pass | 25,523 | 3,249 | -87% | 1 | 1 | 0% | 4,918 | 835 | -83% | 0 | 0 | — |
case-10 | fail→pass | 12,870 | 3,190 | -75% | 1 | 1 | 0% | 2,439 | 779 | -68% | 0 | 0 | — |
case-11 | fail→pass | 9,178 | 3,435 | -63% | 1 | 1 | 0% | 1,349 | 879 | -35% | 0 | 0 | — |
case-12 | fail→pass | 21,433 | 4,272 | -80% | 1 | 1 | 0% | 1,142 | 1,026 | -10% | 0 | 0 | — |
case-13 | fail→pass | 6,514 | 2,805 | -57% | 1 | 1 | 0% | 1,267 | 680 | -46% | 0 | 0 | — |
case-24 | pass→fail | 18,811 | 34,359 | +83% | 1 | 1 | 0% | 3,551 | 5,940 | +67% | 0 | 0 | — |
case-14 | fail→pass | 43,414 | 5,359 | -88% | 1 | 1 | 0% | 8,225 | 1,030 | -87% | 0 | 0 | — |
case-15 | fail→pass | 3,198 | 3,771 | +18% | 1 | 1 | 0% | 134 | 721 | +438% | 0 | 0 | — |
case-16 | fail→pass | 11,179 | 3,296 | -71% | 1 | 1 | 0% | 2,291 | 880 | -62% | 0 | 0 | — |
case-17 | fail→pass | 6,634 | 2,935 | -56% | 1 | 1 | 0% | 1,292 | 803 | -38% | 0 | 0 | — |
case-18 | fail→fail | 10,531 | 8,939 | -15% | 1 | 1 | 0% | 1,867 | 779 | -58% | 0 | 0 | — |
case-19 | fail→pass | 12,676 | 3,632 | -71% | 1 | 1 | 0% | 2,534 | 894 | -65% | 0 | 0 | — |
case-20 | fail→pass | 5,181 | 3,204 | -38% | 1 | 1 | 0% | 1,013 | 831 | -18% | 0 | 0 | — |
case-21 | fail→pass | 34,983 | 3,452 | -90% | 1 | 1 | 0% | 6,041 | 866 | -86% | 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. 24 cases were attempted, and 21 counted toward the lift figure. The other 3 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 +79 percentage points is the difference between those two pass rates over the 21 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
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 | +36% |
Other measured skills in the registry, with their headline benchmark lift.