Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Draft a LaTeX research paper from all previous stage outputs
.claude/skills/openraiser-nanoresearch-writing/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-06 | ✗→✓ | ▲ Improved | -42% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -57% | 0% |
| case-13 | ✗→✓ | ▲ Improved | -58% | 0% |
| case-17 | ✗→✓ | ▲ Improved | -35% | 0% |
| case-18 | ✗→✓ | ▲ Improved | -35% | 0% |
Take all previous outputs (ideation, planning, experiment results) and produce a complete LaTeX paper draft with figures, tables, and bibliography.
generate_latex: Generate and assemble LaTeX source files for each paper sectioncompile_pdf: Compile the LaTeX source into a PDF documentgenerate_figure: Produce publication-quality figures from experiment resultsideation_output: Path to papers/ideation_output.json from the ideation skillexperiment_blueprint: Path to papers/experiment_blueprint.json from the planning skillexperiment_results: Path to experiments/ directory containing code and results from the experiment skillgenerate_figuregenerate_latexcompile_pdf and verify the outputProduces papers/draft/ directory containing:
main.tex: Complete LaTeX source of the paperreferences.bib: Bibliography file with all citationsfigures/: Generated figures in PDF or PNG formattables/: LaTeX table source filesmain.pdf: Compiled PDF of the paper draft| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-07 | pass→pass | 11,977 | 4,258 | -64% | 1 | 1 | 0% | 1,974 | 1,163 | -41% | 0 | 0 | — |
case-01 | fail→fail | 4,234 | 4,748 | +12% | 1 | 1 | 0% | 208 | 661 | +218% | 0 | 0 | — |
case-08 | pass→pass | 7,924 | 4,564 | -42% | 1 | 1 | 0% | 1,343 | 1,200 | -11% | 0 | 0 | — |
case-02 | fail→fail | 29,098 | 3,327 | -89% | 1 | 1 | 0% | 6,190 | 577 | -91% | 0 | 0 | — |
case-03 | pass→pass | 2,733 | 2,524 | -8% | 1 | 1 | 0% | 439 | 779 | +77% | 0 | 0 | — |
case-04 | pass→fail | 3,777 | 11,341 | +200% | 1 | 1 | 0% | 588 | 1,016 | +73% | 0 | 0 | — |
case-05 | pass→pass | 21,829 | 21,011 | -4% | 1 | 1 | 0% | 3,700 | 4,244 | +15% | 0 | 0 | — |
case-06 | fail→pass | 12,179 | 4,649 | -62% | 1 | 1 | 0% | 2,206 | 1,280 | -42% | 0 | 0 | — |
case-09 | pass→pass | 10,348 | 7,146 | -31% | 1 | 1 | 0% | 1,877 | 1,690 | -10% | 0 | 0 | — |
case-10 | fail→pass | 10,913 | 2,094 | -81% | 1 | 1 | 0% | 1,936 | 824 | -57% | 0 | 0 | — |
case-11 | pass→fail | 10,978 | 7,347 | -33% | 1 | 1 | 0% | 1,788 | 1,710 | -4% | 0 | 0 | — |
case-12 | pass→pass | 12,966 | 12,155 | -6% | 1 | 1 | 0% | 2,366 | 2,660 | +12% | 0 | 0 | — |
case-13 | fail→pass | 9,706 | 1,533 | -84% | 1 | 1 | 0% | 1,764 | 733 | -58% | 0 | 0 | — |
case-14 | pass→pass | 13,740 | 6,826 | -50% | 1 | 1 | 0% | 2,513 | 1,721 | -32% | 0 | 0 | — |
case-15 | pass→pass | 9,808 | 5,874 | -40% | 1 | 1 | 0% | 1,762 | 1,308 | -26% | 0 | 0 | — |
case-16 | pass→pass | 9,320 | 2,407 | -74% | 1 | 1 | 0% | 1,798 | 822 | -54% | 0 | 0 | — |
case-17 | fail→pass | 7,054 | 2,434 | -65% | 1 | 1 | 0% | 1,377 | 901 | -35% | 0 | 0 | — |
case-18 | fail→pass | 5,952 | 1,375 | -77% | 1 | 1 | 0% | 1,036 | 678 | -35% | 0 | 0 | — |
case-19 | fail→pass | 8,243 | 1,665 | -80% | 1 | 1 | 0% | 1,623 | 781 | -52% | 0 | 0 | — |
case-20 | fail→fail | 10,263 | 2,614 | -75% | 1 | 1 | 0% | 1,883 | 858 | -54% | 0 | 0 | — |
case-21 | pass→pass | 9,709 | 6,396 | -34% | 1 | 1 | 0% | 1,738 | 1,401 | -19% | 0 | 0 | — |
case-22 | fail→pass | 9,929 | 2,953 | -70% | 1 | 1 | 0% | 1,679 | 946 | -44% | 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 19 counted toward the lift figure. The other 3 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 +23 percentage points is the difference between those two pass rates over the 19 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.