Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →단일 brand의 references/<id>/DESIGN.md를 Apple-tier 깊이로 풀 마이그레이션. 9 quality gate 전부 통과 — §4 canonical schema + ≥2 surface live inspect + Tier 2 cross-check + .verification.md conflict matrix + 검증된 §10-15 philosophy + footer + tests + visual smoke. 'X 마이그레이션', 'X apple 깊이로', 'X 풀 검증', 'omd:migrate stripe' 류에 트리거.
.claude/skills/kwakseongjae-omd-migrate/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | -30% | 0% |
| case-16 | ✗→✓ | ▲ Improved | -25% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 252% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 104% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 127% | 0% |
CORE_V2_CATALOG_WRITE_BLOCKED
Do not mutate web/references/<id>/DESIGN.md, its frontmatter, numbered sections, verification footer, or mirrors. Those files remain readable during the migration window, but creating more legacy bytes would violate the Core v2 dual-read/single-write contract.
For a provider-free assessment, use the Core migration auditor in check mode. Do not adopt or overwrite its staged result. A migration candidate keeps the source DESIGN.md canonical until a graph-backed catalog writer, migrated consumers, evidence binding, and dropped_segments=0 acceptance ship together.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 7,223 | 4,068 | -44% | 1 | 1 | 0% | 1,296 | 902 | -30% | 0 | 0 | — |
case-02 | pass→fail | 16,118 | 6,163 | -62% | 1 | 1 | 0% | 2,717 | 564 | -79% | 0 | 0 | — |
case-03 | fail→fail | 4,304 | 3,448 | -20% | 1 | 1 | 0% | 227 | 846 | +273% | 0 | 0 | — |
case-16 | fail→pass | 4,980 | 2,232 | -55% | 1 | 1 | 0% | 738 | 555 | -25% | 0 | 0 | — |
case-04 | fail→fail | 5,328 | 10,771 | +102% | 1 | 1 | 0% | 403 | 2,002 | +397% | 0 | 0 | — |
case-05 | fail→fail | 6,265 | 6,455 | +3% | 1 | 1 | 0% | 1,166 | 1,334 | +14% | 0 | 0 | — |
case-06 | fail→fail | 3,307 | 5,274 | +59% | 1 | 1 | 0% | 509 | 834 | +64% | 0 | 0 | — |
case-07 | fail→pass | 4,642 | 5,102 | +10% | 1 | 1 | 0% | 299 | 1,051 | +252% | 0 | 0 | — |
case-08 | fail→pass | 5,267 | 2,990 | -43% | 1 | 1 | 0% | 343 | 701 | +104% | 0 | 0 | — |
case-09 | fail→fail | 6,438 | 6,001 | -7% | 1 | 1 | 0% | 380 | 537 | +41% | 0 | 0 | — |
case-10 | fail→pass | 7,402 | 3,805 | -49% | 1 | 1 | 0% | 377 | 855 | +127% | 0 | 0 | — |
case-11 | fail→pass | 5,579 | 3,241 | -42% | 1 | 1 | 0% | 895 | 780 | -13% | 0 | 0 | — |
case-21 | pass→pass | 8,913 | 4,032 | -55% | 1 | 1 | 0% | 1,408 | 886 | -37% | 0 | 0 | — |
case-12 | pass→pass | 5,589 | 3,367 | -40% | 1 | 1 | 0% | 926 | 799 | -14% | 0 | 0 | — |
case-13 | pass→pass | 3,844 | 3,690 | -4% | 1 | 1 | 0% | 571 | 866 | +52% | 0 | 0 | — |
case-14 | fail→pass | 11,269 | 2,405 | -79% | 1 | 1 | 0% | 1,957 | 571 | -71% | 0 | 0 | — |
case-15 | fail→pass | 2,566 | 2,119 | -17% | 1 | 1 | 0% | 329 | 548 | +67% | 0 | 0 | — |
case-17 | fail→pass | 14,572 | 9,264 | -36% | 1 | 1 | 0% | 2,396 | 1,010 | -58% | 0 | 0 | — |
case-18 | fail→pass | 24,345 | 2,058 | -92% | 1 | 1 | 0% | 1,969 | 490 | -75% | 0 | 0 | — |
case-19 | fail→pass | 2,408 | 3,813 | +58% | 1 | 1 | 0% | 337 | 884 | +162% | 0 | 0 | — |
case-20 | fail→pass | 7,630 | 3,547 | -54% | 1 | 1 | 0% | 1,501 | 806 | -46% | 0 | 0 | — |
case-22 | fail→pass | 7,287 | 2,391 | -67% | 1 | 1 | 0% | 379 | 611 | +61% | 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. 22 cases were attempted, and 15 counted toward the lift figure. The other 7 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 +55 percentage points is the difference between those two pass rates over the 15 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 | +50% |
Other measured skills in the registry, with their headline benchmark lift.