Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →7 ocr & translation skills. Trigger: scanning documents, recognizing formulas, translating academic papers. Design: specialized OCR (LaTeX, handwriting) and translation for scholarly content.
.claude/skills/brycewang-stanford-ocr-translate-skills/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-12 | ✗→✓ | ▲ Improved | -30% | 0% |
| case-14 | ✗→✓ | ▲ Improved | -52% | 0% |
| case-19 | ✗→✓ | ▲ Improved | -42% | 0% |
| case-03 | ✓→✗ | ▼ Worse | -74% | 0% |
| case-06 | ✓→✗ | ▼ Worse | -77% | 0% |
Select the skill matching the user's need, then read its SKILL.md.
| Skill | Description | |-------|-------------| | handwriting-recognition-guide | Apply handwriting OCR to digitize historical and archival documents | | latex-ocr-guide | Extract and convert mathematical formulas from images and PDFs to LaTeX code | | latex-translation-guide | Translate LaTeX documents preserving math formulas and structure | | multilingual-research-guide | Strategies for translating academic papers while preserving technical accuracy | | pdf-math-translate-guide | Translate scientific PDFs with preserved math formatting via PDFMathTranslate | | zotero-pdf-translate-guide | Guide to Zotero PDF Translate for multilingual PDF and annotation translation | | zotero-pdf2zh-guide | PDF Chinese translation plugin for Zotero reference manager |
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-09 | pass→pass | 16,833 | 9,921 | -41% | 1 | 1 | 0% | 2,722 | 1,614 | -41% | 0 | 0 | — |
case-01 | fail→fail | 14,809 | 4,630 | -69% | 1 | 1 | 0% | 2,272 | 915 | -60% | 0 | 0 | — |
case-02 | fail→fail | 17,746 | 19,394 | +9% | 1 | 1 | 0% | 2,870 | 682 | -76% | 0 | 0 | — |
case-03 | pass→fail | 14,578 | 4,239 | -71% | 1 | 1 | 0% | 2,571 | 672 | -74% | 0 | 0 | — |
case-04 | pass→pass | 10,584 | 9,960 | -6% | 1 | 1 | 0% | 1,721 | 1,467 | -15% | 0 | 0 | — |
case-05 | pass→pass | 18,261 | 9,274 | -49% | 1 | 1 | 0% | 2,957 | 1,902 | -36% | 0 | 0 | — |
case-06 | pass→fail | 18,369 | 5,025 | -73% | 1 | 1 | 0% | 2,753 | 630 | -77% | 0 | 0 | — |
case-07 | fail→fail | 14,988 | 6,759 | -55% | 1 | 1 | 0% | 2,417 | 852 | -65% | 0 | 0 | — |
case-08 | pass→fail | 11,417 | 5,935 | -48% | 1 | 1 | 0% | 1,841 | 831 | -55% | 0 | 0 | — |
case-10 | pass→fail | 12,219 | 5,886 | -52% | 1 | 1 | 0% | 1,955 | 778 | -60% | 0 | 0 | — |
case-11 | pass→fail | 15,994 | 3,362 | -79% | 1 | 1 | 0% | 2,533 | 583 | -77% | 0 | 0 | — |
case-12 | fail→pass | 14,816 | 9,155 | -38% | 1 | 1 | 0% | 2,270 | 1,592 | -30% | 0 | 0 | — |
case-13 | pass→fail | 21,696 | 4,968 | -77% | 1 | 1 | 0% | 3,275 | 603 | -82% | 0 | 0 | — |
case-18 | pass→fail | 17,650 | 3,961 | -78% | 1 | 1 | 0% | 2,756 | 570 | -79% | 0 | 0 | — |
case-14 | fail→pass | 13,076 | 4,131 | -68% | 1 | 1 | 0% | 2,143 | 1,020 | -52% | 0 | 0 | — |
case-15 | pass→fail | 14,741 | 3,857 | -74% | 1 | 1 | 0% | 2,402 | 657 | -73% | 0 | 0 | — |
case-16 | pass→fail | 14,818 | 3,950 | -73% | 1 | 1 | 0% | 2,215 | 619 | -72% | 0 | 0 | — |
case-17 | pass→pass | 10,610 | 8,735 | -18% | 1 | 1 | 0% | 1,538 | 1,066 | -31% | 0 | 0 | — |
case-19 | fail→pass | 14,833 | 5,732 | -61% | 1 | 1 | 0% | 2,208 | 1,287 | -42% | 0 | 0 | — |
case-20 | pass→pass | 15,246 | 45,309 | +197% | 1 | 1 | 0% | 2,233 | 2,513 | +13% | 0 | 0 | — |
case-21 | pass→pass | 12,018 | 9,591 | -20% | 1 | 1 | 0% | 2,078 | 2,020 | -3% | 0 | 0 | — |
case-22 | pass→pass | 21,400 | 16,019 | -25% | 1 | 1 | 0% | 4,374 | 3,314 | -24% | 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 11 counted toward the lift figure. The other 11 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 -27 percentage points is the difference between those two pass rates over the 11 comparable cases. 10 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.