Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Treat manuscripts as software: version control, reproducible builds, figure pipelines, CI, and structured repo layout. Helps teams avoid 'final_v7' chaos and ensures submission-ready artifacts.
.claude/skills/foryourhealth111-pixel-manuscript-as-code/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 19% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 8% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 4% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -2% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -11% | 0% |
把“写论文”升级成“交付可复现的出版工程”：
适用：
见 templates/repo-structure.md。核心思想：
manuscript/：只放正文与引用（source-of-truth）figures/：每张图一个目录（source + out）build/：所有生成物（可删除，可再生）submission/ 与 revision/：投稿与返修阶段产物figures/fig-02/src/plot.pyfigures/fig-02/out/fig-02.pdf最少要求：
推荐维护：
submission/submission-manifest.yml（统一记录规格与完成度）manubot/rootstock（论文的 CI/协作写作范式）greenelab/deep-review（评审/返修视角）| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 16,629 | 16,787 | +1% | 1 | 1 | 0% | 2,842 | 3,379 | +19% | 0 | 0 | — |
case-02 | fail→pass | 21,659 | 19,494 | -10% | 1 | 1 | 0% | 3,539 | 3,837 | +8% | 0 | 0 | — |
case-03 | fail→pass | 19,926 | 18,815 | -6% | 1 | 1 | 0% | 3,526 | 3,661 | +4% | 0 | 0 | — |
case-04 | pass→pass | 17,278 | 16,621 | -4% | 1 | 1 | 0% | 2,733 | 3,263 | +19% | 0 | 0 | — |
case-05 | pass→pass | 15,609 | 15,980 | +2% | 1 | 1 | 0% | 2,659 | 3,317 | +25% | 0 | 0 | — |
case-06 | pass→pass | 18,461 | 20,340 | +10% | 1 | 1 | 0% | 3,170 | 4,180 | +32% | 0 | 0 | — |
case-07 | fail→pass | 17,491 | 13,986 | -20% | 1 | 1 | 0% | 2,845 | 2,784 | -2% | 0 | 0 | — |
case-08 | fail→pass | 18,318 | 14,066 | -23% | 1 | 1 | 0% | 3,120 | 2,792 | -11% | 0 | 0 | — |
case-09 | pass→pass | 16,813 | 13,325 | -21% | 1 | 1 | 0% | 2,728 | 2,679 | -2% | 0 | 0 | — |
case-15 | fail→pass | 17,317 | 12,899 | -26% | 1 | 1 | 0% | 2,391 | 2,394 | +0% | 0 | 0 | — |
case-10 | pass→pass | 17,179 | 16,110 | -6% | 1 | 1 | 0% | 2,646 | 3,203 | +21% | 0 | 0 | — |
case-11 | pass→pass | 16,784 | 17,094 | +2% | 1 | 1 | 0% | 2,883 | 3,348 | +16% | 0 | 0 | — |
case-12 | fail→fail | 18,996 | 16,089 | -15% | 1 | 1 | 0% | 3,054 | 3,019 | -1% | 0 | 0 | — |
case-13 | fail→fail | 19,605 | 19,355 | -1% | 1 | 1 | 0% | 3,271 | 4,026 | +23% | 0 | 0 | — |
case-14 | pass→pass | 19,054 | 15,301 | -20% | 1 | 1 | 0% | 2,845 | 2,940 | +3% | 0 | 0 | — |
case-16 | fail→fail | 14,377 | 10,739 | -25% | 1 | 1 | 0% | 2,647 | 2,387 | -10% | 0 | 0 | — |
case-17 | pass→pass | 16,416 | 13,847 | -16% | 1 | 1 | 0% | 2,870 | 2,736 | -5% | 0 | 0 | — |
case-18 | fail→pass | 14,524 | 9,726 | -33% | 1 | 1 | 0% | 2,567 | 2,059 | -20% | 0 | 0 | — |
case-19 | pass→pass | 19,465 | 19,138 | -2% | 1 | 1 | 0% | 3,260 | 3,555 | +9% | 0 | 0 | — |
case-20 | pass→pass | 16,510 | 17,307 | +5% | 1 | 1 | 0% | 2,885 | 3,598 | +25% | 0 | 0 | — |
case-21 | pass→pass | 14,813 | 11,997 | -19% | 1 | 1 | 0% | 2,297 | 1,720 | -25% | 0 | 0 | — |
case-22 | fail→pass | 15,090 | 11,497 | -24% | 1 | 1 | 0% | 2,398 | 2,346 | -2% | 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 +36 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.