Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use when `paper-review` needs overlap/delta positioning against provided related work. **Trigger**: novelty matrix, prior-work matrix, overlap/delta, 相关工作对比, 新颖性矩阵. **Use when**: `paper-review` 中评估 novelty/positioning，需要把贡献与相关工作逐项对齐并写出差异点证据。
.claude/skills/willoscar-novelty-matrix/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | 43% | 0% |
| case-12 | ✗→✓ | ▲ Improved | 7% | 0% |
| case-16 | ✗→✓ | ▲ Improved | -55% | 0% |
| case-22 | ✗→✓ | ▲ Improved | -19% | 0% |
| case-23 | ✗→✓ | ▲ Improved | -57% | 0% |
Transforms a claim ledger plus related-work surface into a novelty positioning table for paper-review.
Required:
output/CLAIMS.mdOptional:
output/NOVELTY_MATRIX.mdoutput/NOVELTY_MATRIX.tsv (review-novelty-row.v1)Each matrix row must expose:
scripts/run.py should:
Keep matching heuristics and markdown rendering in shared tooling.
claim_id, overlap, delta, and evidence as separate fields| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 8,445 | 3,805 | -55% | 1 | 1 | 0% | 1,239 | 433 | -65% | 0 | 0 | — |
case-02 | fail→fail | 25,774 | 4,438 | -83% | 1 | 1 | 0% | 4,953 | 490 | -90% | 0 | 0 | — |
case-03 | fail→fail | 11,527 | 4,797 | -58% | 1 | 1 | 0% | 1,929 | 489 | -75% | 0 | 0 | — |
case-04 | fail→pass | 25,874 | 7,456 | -71% | 1 | 1 | 0% | 959 | 1,374 | +43% | 0 | 0 | — |
case-05 | pass→pass | 13,096 | 4,635 | -65% | 1 | 1 | 0% | 2,106 | 1,006 | -52% | 0 | 0 | — |
case-06 | pass→fail | 2,620 | 4,942 | +89% | 1 | 1 | 0% | 331 | 471 | +42% | 0 | 0 | — |
case-07 | fail→fail | 5,424 | 2,157 | -60% | 1 | 1 | 0% | 846 | 602 | -29% | 0 | 0 | — |
case-08 | pass→pass | 5,033 | 4,849 | -4% | 1 | 1 | 0% | 789 | 980 | +24% | 0 | 0 | — |
case-09 | fail→fail | 9,806 | 6,510 | -34% | 1 | 1 | 0% | 1,792 | 583 | -67% | 0 | 0 | — |
case-10 | fail→fail | 15,449 | 5,748 | -63% | 1 | 1 | 0% | 2,274 | 618 | -73% | 0 | 0 | — |
case-11 | fail→fail | 16,437 | 6,118 | -63% | 1 | 1 | 0% | 3,021 | 401 | -87% | 0 | 0 | — |
case-12 | fail→pass | 17,986 | 20,933 | +16% | 1 | 1 | 0% | 3,308 | 3,550 | +7% | 0 | 0 | — |
case-13 | pass→fail | 25,880 | 6,039 | -77% | 1 | 1 | 0% | 4,165 | 643 | -85% | 0 | 0 | — |
case-14 | fail→fail | 4,837 | 5,594 | +16% | 1 | 1 | 0% | 678 | 442 | -35% | 0 | 0 | — |
case-15 | fail→fail | 14,537 | 15,075 | +4% | 1 | 1 | 0% | 2,027 | 2,304 | +14% | 0 | 0 | — |
case-16 | fail→pass | 11,768 | 3,691 | -69% | 1 | 1 | 0% | 1,842 | 832 | -55% | 0 | 0 | — |
case-17 | fail→fail | 3,241 | 4,795 | +48% | 1 | 1 | 0% | 290 | 614 | +112% | 0 | 0 | — |
case-18 | pass→pass | 8,520 | 6,406 | -25% | 1 | 1 | 0% | 1,255 | 999 | -20% | 0 | 0 | — |
case-19 | fail→fail | 16,367 | 4,085 | -75% | 1 | 1 | 0% | 2,581 | 382 | -85% | 0 | 0 | — |
case-20 | fail→fail | 4,617 | 4,568 | -1% | 1 | 1 | 0% | 196 | 439 | +124% | 0 | 0 | — |
case-21 | fail→fail | 22,221 | 7,031 | -68% | 1 | 1 | 0% | 3,919 | 711 | -82% | 0 | 0 | — |
case-22 | fail→pass | 11,332 | 7,391 | -35% | 1 | 1 | 0% | 1,615 | 1,305 | -19% | 0 | 0 | — |
case-23 | fail→pass | 9,134 | 2,280 | -75% | 1 | 1 | 0% | 1,364 | 585 | -57% | 0 | 0 | — |
case-24 | fail→pass | 7,196 | 3,204 | -55% | 1 | 1 | 0% | 976 | 828 | -15% | 0 | 0 | — |
case-25 | fail→fail | 4,542 | 2,135 | -53% | 1 | 1 | 0% | 646 | 536 | -17% | 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. 25 cases were attempted, and 12 counted toward the lift figure. The other 13 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 +16 percentage points is the difference between those two pass rates over the 12 comparable cases. 2 cases got worse with the skill loaded, and they are 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.
Other measured skills in the registry, with their headline benchmark lift.