Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →10개 brand 일괄 reference 추가 + 문서 카운트 sync + hyperframes 기반 promo MP4 생성. '10개씩 추천해줘', '한국 IT 10개 추가하고 영상까지', 'batch-launch', 'X 카테고리에서 10개 더' 류 트리거. 매 회 reference count +10, gitignored promo video 1편.
.claude/skills/kwakseongjae-omd-batch-launch/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-06 | ✗→✓ | ▲ Improved | -43% | 0% |
| case-01 | ✗→✓ | ▲ Improved | 11% | 0% |
| case-02 | ✗→✓ | ▲ Improved | -39% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 36% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 42% | 0% |
Status: CORE_V2_CATALOG_WRITE_BLOCKED.
Do not add references, mutate counts, start authoring workers, or create promo claims from an unshipped batch. The prior workflow emitted legacy frontmatter and numbered sections, so it is disabled until omd:add-reference has a graph-backed Core v2 writer and all catalog readers accept its package.
Candidate research may be returned in chat without repository writes. Never bypass this gate with legacy YAML, placeholder sentinels, or a mixed-format batch.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-06 | fail→pass | 7,809 | 3,188 | -59% | 1 | 1 | 0% | 1,298 | 736 | -43% | 0 | 0 | — |
case-01 | fail→pass | 4,385 | 3,654 | -17% | 1 | 1 | 0% | 694 | 768 | +11% | 0 | 0 | — |
case-02 | fail→pass | 8,812 | 4,852 | -45% | 1 | 1 | 0% | 1,570 | 958 | -39% | 0 | 0 | — |
case-03 | fail→pass | 3,514 | 3,760 | +7% | 1 | 1 | 0% | 583 | 793 | +36% | 0 | 0 | — |
case-04 | fail→pass | 4,048 | 3,963 | -2% | 1 | 1 | 0% | 587 | 834 | +42% | 0 | 0 | — |
case-05 | fail→pass | 3,084 | 5,910 | +92% | 1 | 1 | 0% | 454 | 717 | +58% | 0 | 0 | — |
case-07 | fail→pass | 8,064 | 3,855 | -52% | 1 | 1 | 0% | 1,366 | 756 | -45% | 0 | 0 | — |
case-08 | fail→pass | 8,808 | 4,996 | -43% | 1 | 1 | 0% | 1,528 | 1,029 | -33% | 0 | 0 | — |
case-09 | fail→pass | 15,596 | 5,687 | -64% | 1 | 1 | 0% | 2,462 | 1,096 | -55% | 0 | 0 | — |
case-10 | fail→pass | 10,249 | 4,749 | -54% | 1 | 1 | 0% | 1,609 | 970 | -40% | 0 | 0 | — |
case-11 | fail→pass | 3,260 | 3,683 | +13% | 1 | 1 | 0% | 460 | 698 | +52% | 0 | 0 | — |
case-12 | pass→pass | 3,731 | 3,847 | +3% | 1 | 1 | 0% | 580 | 757 | +31% | 0 | 0 | — |
case-13 | pass→pass | 15,888 | 14,768 | -7% | 1 | 1 | 0% | 2,655 | 2,433 | -8% | 0 | 0 | — |
case-14 | fail→fail | 4,390 | 4,549 | +4% | 1 | 1 | 0% | 666 | 797 | +20% | 0 | 0 | — |
case-15 | pass→pass | 8,049 | 10,028 | +25% | 1 | 1 | 0% | 1,225 | 1,755 | +43% | 0 | 0 | — |
case-16 | fail→pass | 7,287 | 2,807 | -61% | 1 | 1 | 0% | 1,126 | 671 | -40% | 0 | 0 | — |
case-17 | fail→pass | 8,927 | 4,167 | -53% | 1 | 1 | 0% | 1,702 | 831 | -51% | 0 | 0 | — |
case-18 | pass→pass | 6,192 | 3,490 | -44% | 1 | 1 | 0% | 1,130 | 767 | -32% | 0 | 0 | — |
case-19 | fail→pass | 4,650 | 3,190 | -31% | 1 | 1 | 0% | 797 | 704 | -12% | 0 | 0 | — |
case-20 | fail→pass | 6,600 | 4,344 | -34% | 1 | 1 | 0% | 1,117 | 884 | -21% | 0 | 0 | — |
case-21 | pass→pass | 7,001 | 3,464 | -51% | 1 | 1 | 0% | 1,098 | 746 | -32% | 0 | 0 | — |
case-22 | fail→fail | 5,701 | 4,207 | -26% | 1 | 1 | 0% | 905 | 830 | -8% | 0 | 0 | — |
case-23 | pass→pass | 32,765 | 26,007 | -21% | 1 | 1 | 0% | 1,421 | 1,094 | -23% | 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. The headline lift of +65 percentage points is the difference between those two pass rates over the 23 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 | +67% |
Other measured skills in the registry, with their headline benchmark lift.