Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Refine a Stage2 NeurIPS rebuttal draft into polished, reviewer-facing prose in the author's style; preserve factual grounding, optionally prepend a courteous opening phrase, and normalize tables/code/formulas into Markdown.
.claude/skills/runtsang-stage2-neurips-refine/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-06 | ✗→✓ | ▲ Improved | -55% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -11% | 0% |
| case-11 | ✗→✓ | ▲ Improved | -22% | 0% |
| case-12 | ✗→✓ | ▲ Improved | -24% | 0% |
| case-17 | ✗→✓ | ▲ Improved | -12% | 0% |
Apply the shared template first:
skills/stage2/template/SKILL.mdThen apply only the NeurIPS-specific overrides below.
No additional structural override is required beyond conference labeling. All writing constraints, opener rules, normalization rules, and hard constraints are inherited from the base template.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-02 | fail→fail | 21,710 | 10,636 | -51% | 1 | 1 | 0% | 2,307 | 1,947 | -16% | 0 | 0 | — |
case-01 | fail→fail | 13,299 | 26,521 | +99% | 1 | 1 | 0% | 1,750 | 2,135 | +22% | 0 | 0 | — |
case-03 | fail→fail | 1,775 | 4,385 | +147% | 1 | 1 | 0% | 197 | 733 | +272% | 0 | 0 | — |
case-04 | pass→pass | 57,938 | 56,175 | -3% | 1 | 1 | 0% | 8,221 | 8,388 | +2% | 0 | 0 | — |
case-05 | pass→pass | 20,779 | 26,959 | +30% | 1 | 1 | 0% | 2,700 | 2,884 | +7% | 0 | 0 | — |
case-06 | fail→pass | 8,709 | 2,375 | -73% | 1 | 1 | 0% | 1,099 | 491 | -55% | 0 | 0 | — |
case-07 | pass→pass | 11,287 | 8,137 | -28% | 1 | 1 | 0% | 1,606 | 1,553 | -3% | 0 | 0 | — |
case-08 | fail→fail | 11,522 | 7,418 | -36% | 1 | 1 | 0% | 1,701 | 1,140 | -33% | 0 | 0 | — |
case-09 | fail→fail | 12,655 | 10,427 | -18% | 1 | 1 | 0% | 1,966 | 1,605 | -18% | 0 | 0 | — |
case-10 | fail→pass | 15,313 | 10,180 | -34% | 1 | 1 | 0% | 1,979 | 1,768 | -11% | 0 | 0 | — |
case-11 | fail→pass | 8,385 | 6,798 | -19% | 1 | 1 | 0% | 1,332 | 1,038 | -22% | 0 | 0 | — |
case-12 | fail→pass | 7,452 | 4,129 | -45% | 1 | 1 | 0% | 957 | 729 | -24% | 0 | 0 | — |
case-13 | fail→fail | 10,300 | 10,594 | +3% | 1 | 1 | 0% | 1,757 | 1,289 | -27% | 0 | 0 | — |
case-14 | pass→pass | 9,113 | 12,118 | +33% | 1 | 1 | 0% | 1,185 | 1,627 | +37% | 0 | 0 | — |
case-15 | fail→fail | 8,924 | 10,498 | +18% | 1 | 1 | 0% | 1,437 | 1,887 | +31% | 0 | 0 | — |
case-16 | fail→fail | 8,449 | 12,026 | +42% | 1 | 1 | 0% | 1,173 | 1,728 | +47% | 0 | 0 | — |
case-17 | fail→pass | 13,389 | 10,194 | -24% | 1 | 1 | 0% | 1,849 | 1,619 | -12% | 0 | 0 | — |
case-18 | pass→pass | 11,276 | 9,126 | -19% | 1 | 1 | 0% | 1,602 | 1,526 | -5% | 0 | 0 | — |
case-19 | fail→pass | 13,687 | 7,632 | -44% | 1 | 1 | 0% | 1,994 | 1,186 | -41% | 0 | 0 | — |
case-20 | pass→pass | 10,835 | 6,364 | -41% | 1 | 1 | 0% | 1,680 | 1,216 | -28% | 0 | 0 | — |
case-21 | fail→pass | 8,253 | 4,421 | -46% | 1 | 1 | 0% | 1,312 | 768 | -41% | 0 | 0 | — |
case-22 | fail→fail | 21,162 | 8,680 | -59% | 1 | 1 | 0% | 1,787 | 1,373 | -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. 22 cases were attempted. The headline lift of +32 percentage points is the difference between those two pass rates over the 22 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.
Other measured skills in the registry, with their headline benchmark lift.