Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Write research/technical reports with strong structure + figure standards. Supports Markdown/HTML/PDF outputs (Quarto optional), executive summary, methods, results, discussion, and reproducibility appendix.
.claude/skills/foryourhealth111-pixel-scientific-reporting/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-06 | ✗→✓ | ▲ Improved | -38% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -3% | 0% |
| case-21 | ✗→✓ | ▲ Improved | 40% | 0% |
| case-09 | ✓→✓ | ＝ Same ✓ | 88% | 0% |
| case-14 | ✓→✓ | ＝ Same ✓ | -2% | 0% |
很多“报告写作失败”不是写不出来，而是：
docs-media（被 PDF/Word 关键词抢走）这个 skill 的职责是：把报告当成可交付工程，并把图表与叙事纳入同一规范。
科研报告、技术报告、项目报告、实验报告、分析报告、HTML 报告、PDF 报告technical report、research report、HTML report、Quarto、RMarkdown不适用：
docs-media/pdf/docx/markitdown1) 受众：老板/评审/客户/组会同学（决定写作风格与解释深度） 2) 目的：决策/复现/汇报/归档（决定是否必须包含 appendix） 3) 输出格式：Markdown / HTML / PDF（可多选） 4) 数据与图：是否已有数据、已有图、或需要从数据生成图
推荐输出为（可按项目名建子目录）：
reports/<topic>/report.md（source-of-truth）reports/<topic>/report.html（可选）reports/<topic>/report.pdf（可选）reports/<topic>/figures/（报告引用的图，来源于 figures/**/out）reports/<topic>/appendix/（方法、参数、环境、额外图表）模板：templates/report-skeleton.md
报告必须包含（即使简写）： 1) Executive Summary（结论先行） 2) Context（问题与约束） 3) Methods（可复现：数据/流程/参数） 4) Results（以图为骨架） 5) Discussion（解释意义、局限、建议） 6) Appendix（可选但推荐：环境/补充图/表）
scientific-visualization（publication-quality）scientific-schematics（流程/机制/系统图）scientific-writing（段落化、避免 bullet 作为最终稿）如果要 HTML：
如果要 PDF：
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-09 | pass→pass | 10,738 | 16,601 | +55% | 1 | 1 | 0% | 1,900 | 3,579 | +88% | 0 | 0 | — |
case-14 | pass→pass | 23,474 | 18,206 | -22% | 1 | 1 | 0% | 3,818 | 3,740 | -2% | 0 | 0 | — |
case-15 | pass→pass | 10,265 | 7,074 | -31% | 1 | 1 | 0% | 1,509 | 1,801 | +19% | 0 | 0 | — |
case-04 | pass→pass | 11,967 | 10,052 | -16% | 1 | 1 | 0% | 1,973 | 2,333 | +18% | 0 | 0 | — |
case-01 | pass→pass | 13,408 | 9,025 | -33% | 1 | 1 | 0% | 2,494 | 2,510 | +1% | 0 | 0 | — |
case-02 | pass→pass | 7,624 | 5,799 | -24% | 1 | 1 | 0% | 1,387 | 1,910 | +38% | 0 | 0 | — |
case-03 | pass→pass | 10,714 | 8,016 | -25% | 1 | 1 | 0% | 2,004 | 2,346 | +17% | 0 | 0 | — |
case-05 | pass→pass | 11,929 | 9,771 | -18% | 1 | 1 | 0% | 1,924 | 2,330 | +21% | 0 | 0 | — |
case-06 | fail→pass | 25,538 | 3,593 | -86% | 1 | 1 | 0% | 2,281 | 1,418 | -38% | 0 | 0 | — |
case-07 | pass→pass | 6,863 | 3,532 | -49% | 1 | 1 | 0% | 1,064 | 1,392 | +31% | 0 | 0 | — |
case-08 | fail→pass | 10,638 | 6,300 | -41% | 1 | 1 | 0% | 1,918 | 1,869 | -3% | 0 | 0 | — |
case-10 | pass→pass | 14,993 | 14,702 | -2% | 1 | 1 | 0% | 2,480 | 3,146 | +27% | 0 | 0 | — |
case-11 | pass→pass | 16,223 | 13,543 | -17% | 1 | 1 | 0% | 2,781 | 2,827 | +2% | 0 | 0 | — |
case-12 | pass→pass | 14,977 | 14,346 | -4% | 1 | 1 | 0% | 2,589 | 3,392 | +31% | 0 | 0 | — |
case-13 | pass→pass | 11,233 | 10,731 | -4% | 1 | 1 | 0% | 1,939 | 2,556 | +32% | 0 | 0 | — |
case-16 | pass→pass | 12,005 | 6,182 | -49% | 1 | 1 | 0% | 1,934 | 1,835 | -5% | 0 | 0 | — |
case-17 | pass→pass | 14,154 | 13,803 | -2% | 1 | 1 | 0% | 2,441 | 3,273 | +34% | 0 | 0 | — |
case-18 | pass→pass | 13,080 | 11,591 | -11% | 1 | 1 | 0% | 2,071 | 2,690 | +30% | 0 | 0 | — |
case-19 | pass→pass | 13,205 | 11,575 | -12% | 1 | 1 | 0% | 2,068 | 2,717 | +31% | 0 | 0 | — |
case-20 | pass→pass | 14,932 | 14,671 | -2% | 1 | 1 | 0% | 2,418 | 3,223 | +33% | 0 | 0 | — |
case-21 | fail→pass | 16,055 | 15,705 | -2% | 1 | 1 | 0% | 2,371 | 3,309 | +40% | 0 | 0 | — |
case-22 | pass→pass | 15,234 | 14,485 | -5% | 1 | 1 | 0% | 2,443 | 3,156 | +29% | 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 +14 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.