Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Convert PDF, Office, HTML, data, media, ZIP to Markdown.
.claude/skills/notque-markdown-converter/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | -35% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 39% | 0% |
| case-01 | ✗→✓ | ▲ Improved | -39% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -12% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -48% | 0% |
Convert a file to Markdown with markitdown, zero install:
bashuvx 'markitdown[all]' input.pdf -o output.md # to file uvx 'markitdown[all]' input.docx # to stdout cat blob | uvx 'markitdown[all]' -x .pdf # stdin, with extension hint
When uvx is missing, run pipx run 'markitdown[all]' … with the same arguments. First run downloads dependencies; later runs hit the cache. Output preserves headings, tables, lists, and links.
For video transcripts, use the video-transcript skill.
| Input | Notes | |---|---| | PDF, .docx, .pptx, .xlsx, .xls | Document structure preserved | | HTML, CSV, JSON, XML | Structured Markdown | | Images | EXIF metadata + OCR text | | Audio | EXIF metadata + speech transcription | | ZIP, EPub | Iterates contents, converts each |
| Flag | Effect | |---|---| | -o FILE | Write output to FILE | | -x .EXT | Extension hint for stdin input | | -m MIME | MIME-type hint | | -c CHARSET | Charset hint, e.g. UTF-8 |
Cause: page is an image; the base extractor reads text layers only. Solution: render pages to images (pdftoppm), then convert the images so OCR runs.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-02 | fail→pass | 9,061 | 3,395 | -63% | 1 | 1 | 0% | 1,543 | 1,005 | -35% | 0 | 0 | — |
case-03 | fail→pass | 4,633 | 3,006 | -35% | 1 | 1 | 0% | 598 | 833 | +39% | 0 | 0 | — |
case-01 | fail→pass | 9,090 | 4,025 | -56% | 1 | 1 | 0% | 1,625 | 997 | -39% | 0 | 0 | — |
case-04 | fail→pass | 4,369 | 2,147 | -51% | 1 | 1 | 0% | 718 | 632 | -12% | 0 | 0 | — |
case-05 | fail→pass | 7,279 | 2,208 | -70% | 1 | 1 | 0% | 1,325 | 695 | -48% | 0 | 0 | — |
case-06 | fail→pass | 11,801 | 2,576 | -78% | 1 | 1 | 0% | 1,858 | 723 | -61% | 0 | 0 | — |
case-07 | pass→pass | 3,750 | 1,638 | -56% | 1 | 1 | 0% | 650 | 599 | -8% | 0 | 0 | — |
case-08 | fail→pass | 13,498 | 1,929 | -86% | 1 | 1 | 0% | 2,265 | 651 | -71% | 0 | 0 | — |
case-09 | fail→pass | 9,408 | 3,396 | -64% | 1 | 1 | 0% | 1,632 | 678 | -58% | 0 | 0 | — |
case-10 | fail→pass | 9,286 | 3,381 | -64% | 1 | 1 | 0% | 1,529 | 849 | -44% | 0 | 0 | — |
case-11 | fail→pass | 5,848 | 2,304 | -61% | 1 | 1 | 0% | 1,000 | 712 | -29% | 0 | 0 | — |
case-12 | fail→pass | 9,211 | 2,457 | -73% | 1 | 1 | 0% | 1,516 | 771 | -49% | 0 | 0 | — |
case-13 | fail→pass | 15,401 | 3,987 | -74% | 1 | 1 | 0% | 2,807 | 995 | -65% | 0 | 0 | — |
case-14 | fail→pass | 10,500 | 2,861 | -73% | 1 | 1 | 0% | 1,777 | 855 | -52% | 0 | 0 | — |
case-15 | fail→pass | 9,562 | 2,589 | -73% | 1 | 1 | 0% | 1,547 | 743 | -52% | 0 | 0 | — |
case-16 | fail→pass | 10,903 | 2,315 | -79% | 1 | 1 | 0% | 1,895 | 751 | -60% | 0 | 0 | — |
case-17 | fail→pass | 13,018 | 3,688 | -72% | 1 | 1 | 0% | 2,194 | 979 | -55% | 0 | 0 | — |
case-18 | fail→pass | 5,368 | 1,683 | -69% | 1 | 1 | 0% | 997 | 651 | -35% | 0 | 0 | — |
case-19 | pass→pass | 6,324 | 2,217 | -65% | 1 | 1 | 0% | 1,102 | 631 | -43% | 0 | 0 | — |
case-20 | pass→pass | 10,159 | 7,292 | -28% | 1 | 1 | 0% | 1,704 | 1,477 | -13% | 0 | 0 | — |
case-21 | pass→pass | 7,870 | 6,240 | -21% | 1 | 1 | 0% | 1,297 | 1,422 | +10% | 0 | 0 | — |
case-22 | pass→pass | 13,879 | 11,085 | -20% | 1 | 1 | 0% | 2,568 | 2,356 | -8% | 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 +77 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.