Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Read, extract, split, merge, rotate, watermark, fill, OCR, or create PDF files with verification of page counts and text extraction.
.claude/skills/hmbown-pdf/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | -18% | 0% |
| case-09 | ✗→✓ | ▲ Improved | -57% | 0% |
| case-01 | ✓→✗ | ▼ Worse | -80% | 0% |
| case-03 | ✓→✗ | ▼ Worse | -72% | 0% |
| case-07 | ✓→✗ | ▼ Worse | -12% | 0% |
Use this skill for any task where a PDF is the primary input or output.
watermark, redact, fill forms, encrypt/decrypt, or create.
File tool (action: "read") for basic text extraction from PDFspdftotext, pdfinfo, qpdf, or mutool when installedpypdf, pdfplumber, PyMuPDF, orreportlab when available
quality may affect accuracy.
possible, and file size. For redaction, confirm removed text is not extractable from the output.
Ask before installing dependencies or running OCR over large documents. Do not represent a visually scanned PDF as fully accurate text unless OCR quality has been checked.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-06 | fail→fail | 2,377 | 4,776 | +101% | 1 | 1 | 0% | 352 | 451 | +28% | 0 | 0 | — |
case-01 | pass→fail | 24,581 | 7,256 | -70% | 1 | 1 | 0% | 3,969 | 785 | -80% | 0 | 0 | — |
case-02 | fail→fail | 6,200 | 5,471 | -12% | 1 | 1 | 0% | 412 | 539 | +31% | 0 | 0 | — |
case-03 | pass→fail | 9,996 | 5,123 | -49% | 1 | 1 | 0% | 1,912 | 535 | -72% | 0 | 0 | — |
case-04 | fail→pass | 7,093 | 8,456 | +19% | 1 | 1 | 0% | 1,316 | 1,075 | -18% | 0 | 0 | — |
case-05 | fail→fail | 4,138 | 6,418 | +55% | 1 | 1 | 0% | 499 | 685 | +37% | 0 | 0 | — |
case-07 | pass→fail | 3,396 | 5,270 | +55% | 1 | 1 | 0% | 495 | 435 | -12% | 0 | 0 | — |
case-08 | fail→fail | 28,428 | 5,382 | -81% | 1 | 1 | 0% | 6,169 | 524 | -92% | 0 | 0 | — |
case-09 | fail→pass | 7,544 | 2,180 | -71% | 1 | 1 | 0% | 1,428 | 616 | -57% | 0 | 0 | — |
case-10 | fail→fail | 16,776 | 4,939 | -71% | 1 | 1 | 0% | 3,400 | 443 | -87% | 0 | 0 | — |
case-16 | pass→pass | 19,148 | 11,518 | -40% | 1 | 1 | 0% | 4,489 | 3,067 | -32% | 0 | 0 | — |
case-11 | pass→fail | 5,762 | 9,635 | +67% | 1 | 1 | 0% | 1,036 | 1,174 | +13% | 0 | 0 | — |
case-12 | fail→fail | 9,635 | 4,536 | -53% | 1 | 1 | 0% | 1,917 | 580 | -70% | 0 | 0 | — |
case-13 | pass→fail | 5,642 | 5,950 | +5% | 1 | 1 | 0% | 1,076 | 524 | -51% | 0 | 0 | — |
case-14 | pass→fail | 7,571 | 5,293 | -30% | 1 | 1 | 0% | 1,765 | 521 | -70% | 0 | 0 | — |
case-15 | pass→fail | 10,958 | 5,388 | -51% | 1 | 1 | 0% | 2,066 | 637 | -69% | 0 | 0 | — |
case-17 | fail→fail | 5,513 | 9,219 | +67% | 1 | 1 | 0% | 1,031 | 641 | -38% | 0 | 0 | — |
case-18 | fail→fail | 4,525 | 6,616 | +46% | 1 | 1 | 0% | 1,074 | 551 | -49% | 0 | 0 | — |
case-19 | fail→fail | 6,746 | 4,013 | -41% | 1 | 1 | 0% | 1,225 | 528 | -57% | 0 | 0 | — |
case-20 | pass→pass | 13,777 | 14,250 | +3% | 1 | 1 | 0% | 2,883 | 3,196 | +11% | 0 | 0 | — |
case-21 | pass→pass | 9,323 | 6,133 | -34% | 1 | 1 | 0% | 1,672 | 1,384 | -17% | 0 | 0 | — |
case-22 | pass→fail | 6,857 | 4,472 | -35% | 1 | 1 | 0% | 1,421 | 1,194 | -16% | 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 6 counted toward the lift figure. The other 16 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 6 comparable cases. 8 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.
| Model | Method | Date | Lift |
|---|---|---|---|
| gemini-3.6-flash | verified | 8/7/2026 | +9% |
Other measured skills in the registry, with their headline benchmark lift.