Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Break down full NeurIPS reviewer responses into structured rebuttal units. Use when input contains NeurIPS reviewer fields (Summary, Strengths and Weaknesses, Questions, Limitations) and numeric scores (Rating, Confidence, Quality, Clarity, Significance, Originality). Splits questions and limitations into granular response items while preserving original wording for quoted issues.
.claude/skills/runtsang-stage1-neurips-breakdown/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | 113% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 56% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 221% | 0% |
| case-12 | ✗→✓ | ▲ Improved | 199% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 7% | 0% |
Apply the shared template first:
skills/stage1/template/SKILL.mdThen apply only the NeurIPS-specific overrides below.
Extract exactly these six keys (numbers only):
rating <- NeurIPS Rating / Overallconfidence <- NeurIPS Confidencequality <- NeurIPS Qualityclarity <- NeurIPS Claritysignificance <- NeurIPS Significanceoriginality <- NeurIPS OriginalityScores may appear as 3: good or 5: Accept — extract the leading number only.
summary <- Summarystrength <- Strengths And WeaknessesQuestions (label as weakness)Limitations (label as question)# Stage1 NeurIPS Breakdown## Scores must contain exactly the six keys above.Limitations field sometimes contains a standard phrase ("The authors thoroughly discussed the limitations of their work.") — preserve it verbatim rather than splitting into atomic issues if it is a single affirmative statement.4: excellent) — extract only the numeric portion.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-21 | pass→pass | 8,505 | 7,051 | -17% | 1 | 1 | 0% | 1,427 | 1,457 | +2% | 0 | 0 | — |
case-01 | fail→fail | 37,874 | 35,304 | -7% | 1 | 1 | 0% | 1,045 | 1,064 | +2% | 0 | 0 | — |
case-02 | fail→pass | 36,402 | 39,812 | +9% | 1 | 1 | 0% | 977 | 2,085 | +113% | 0 | 0 | — |
case-03 | fail→pass | 6,151 | 37,114 | +503% | 1 | 1 | 0% | 1,080 | 1,685 | +56% | 0 | 0 | — |
case-04 | pass→fail | 10,055 | 11,634 | +16% | 1 | 1 | 0% | 1,854 | 2,610 | +41% | 0 | 0 | — |
case-05 | pass→fail | 41,852 | 51,647 | +23% | 1 | 1 | 0% | 2,040 | 3,241 | +59% | 0 | 0 | — |
case-06 | pass→pass | 11,902 | 39,768 | +234% | 1 | 1 | 0% | 1,805 | 1,855 | +3% | 0 | 0 | — |
case-07 | fail→pass | 3,112 | 8,045 | +159% | 1 | 1 | 0% | 641 | 2,060 | +221% | 0 | 0 | — |
case-12 | fail→pass | 2,998 | 5,078 | +69% | 1 | 1 | 0% | 457 | 1,366 | +199% | 0 | 0 | — |
case-08 | fail→pass | 6,008 | 4,473 | -26% | 1 | 1 | 0% | 1,116 | 1,199 | +7% | 0 | 0 | — |
case-09 | pass→pass | 3,456 | 2,871 | -17% | 1 | 1 | 0% | 710 | 946 | +33% | 0 | 0 | — |
case-10 | pass→pass | 32,242 | 3,090 | -90% | 1 | 1 | 0% | 405 | 815 | +101% | 0 | 0 | — |
case-11 | pass→pass | 4,973 | 3,073 | -38% | 1 | 1 | 0% | 921 | 960 | +4% | 0 | 0 | — |
case-13 | pass→fail | 1,952 | 4,651 | +138% | 1 | 1 | 0% | 338 | 1,227 | +263% | 0 | 0 | — |
case-14 | fail→pass | 5,808 | 4,101 | -29% | 1 | 1 | 0% | 1,005 | 1,192 | +19% | 0 | 0 | — |
case-15 | fail→pass | 4,687 | 4,923 | +5% | 1 | 1 | 0% | 809 | 1,231 | +52% | 0 | 0 | — |
case-16 | fail→pass | 4,313 | 4,372 | +1% | 1 | 1 | 0% | 799 | 1,284 | +61% | 0 | 0 | — |
case-17 | pass→pass | 10,563 | 3,604 | -66% | 1 | 1 | 0% | 1,624 | 941 | -42% | 0 | 0 | — |
case-18 | fail→pass | 5,861 | 3,586 | -39% | 1 | 1 | 0% | 1,112 | 905 | -19% | 0 | 0 | — |
case-19 | pass→pass | 4,318 | 3,444 | -20% | 1 | 1 | 0% | 869 | 862 | -1% | 0 | 0 | — |
case-20 | fail→pass | 4,726 | 6,135 | +30% | 1 | 1 | 0% | 994 | 1,657 | +67% | 0 | 0 | — |
case-22 | fail→pass | 4,332 | 4,602 | +6% | 1 | 1 | 0% | 740 | 1,167 | +58% | 0 | 0 | — |
case-23 | fail→pass | 13,556 | 7,222 | -47% | 1 | 1 | 0% | 2,127 | 1,678 | -21% | 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 +39 percentage points is the difference between those two pass rates over the 23 comparable cases. 3 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.