Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Fills PDF form fields from structured JSON data and writes a flattened output file. Use when the user asks to fill, complete, or populate a PDF form programmatically. Do not use for scanned or image-only PDFs, and not for creating new PDF layouts.
.claude/skills/sayed3li97-pdf-form-filler/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-07 | ✗→✓ | ▲ Improved | -38% | 0% |
| case-11 | ✗→✓ | ▲ Improved | -54% | 0% |
| case-16 | ✗→✓ | ▲ Improved | -10% | 0% |
| case-15 | ✗→✓ | ▲ Improved | -49% | 0% |
| case-17 | ✗→✓ | ▲ Improved | -44% | 0% |
Fill AcroForm fields in an existing PDF from a JSON mapping and produce a flattened copy. Read field mapping reference when the form uses radio groups or nested field names.
python scripts/fill.py --list input.pdf.python scripts/fill.py input.pdf mapping.json out.pdf.--list out.pdf and confirm everyrequired field is populated; if any field is empty, fix the mapping and repeat until the validation passes.
bashpython scripts/fill.py taxform.pdf mapping.json filled.pdf
json{ "applicant_name": "Jane Doe", "filing_year_choice": "single" }
scripts/fill.py only reads the input PDF and writes the named output file; it touches nothing else on disk and never makes network calls. The agent should execute it (not merely read it). Arguments: input path, JSON mapping path, output path; --list prints field names instead of writing.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-02 | fail→fail | 17,986 | 4,760 | -74% | 1 | 1 | 0% | 3,711 | 603 | -84% | 0 | 0 | — |
case-01 | fail→fail | 8,462 | 4,847 | -43% | 1 | 1 | 0% | 533 | 592 | +11% | 0 | 0 | — |
case-03 | fail→fail | 15,660 | 4,510 | -71% | 1 | 1 | 0% | 2,648 | 587 | -78% | 0 | 0 | — |
case-04 | pass→pass | 13,383 | 6,319 | -53% | 1 | 1 | 0% | 2,114 | 1,517 | -28% | 0 | 0 | — |
case-05 | pass→pass | 12,017 | 9,173 | -24% | 1 | 1 | 0% | 2,152 | 1,928 | -10% | 0 | 0 | — |
case-06 | pass→pass | 11,963 | 9,496 | -21% | 1 | 1 | 0% | 2,025 | 1,756 | -13% | 0 | 0 | — |
case-07 | fail→pass | 7,730 | 3,836 | -50% | 1 | 1 | 0% | 1,577 | 979 | -38% | 0 | 0 | — |
case-08 | fail→fail | 12,704 | 16,445 | +29% | 1 | 1 | 0% | 2,019 | 2,240 | +11% | 0 | 0 | — |
case-09 | pass→fail | 7,596 | 4,845 | -36% | 1 | 1 | 0% | 1,445 | 597 | -59% | 0 | 0 | — |
case-10 | pass→pass | 24,581 | 7,985 | -68% | 1 | 1 | 0% | 2,353 | 1,897 | -19% | 0 | 0 | — |
case-11 | fail→pass | 10,806 | 2,424 | -78% | 1 | 1 | 0% | 1,700 | 788 | -54% | 0 | 0 | — |
case-16 | fail→pass | 15,547 | 2,761 | -82% | 1 | 1 | 0% | 1,026 | 927 | -10% | 0 | 0 | — |
case-12 | fail→fail | 9,774 | 2,395 | -75% | 1 | 1 | 0% | 2,150 | 658 | -69% | 0 | 0 | — |
case-13 | pass→pass | 10,269 | 1,816 | -82% | 1 | 1 | 0% | 1,762 | 671 | -62% | 0 | 0 | — |
case-14 | fail→fail | 5,688 | 1,582 | -72% | 1 | 1 | 0% | 1,077 | 583 | -46% | 0 | 0 | — |
case-15 | fail→pass | 11,042 | 3,559 | -68% | 1 | 1 | 0% | 1,884 | 970 | -49% | 0 | 0 | — |
case-17 | fail→pass | 7,211 | 2,061 | -71% | 1 | 1 | 0% | 1,203 | 668 | -44% | 0 | 0 | — |
case-18 | fail→pass | 10,942 | 2,623 | -76% | 1 | 1 | 0% | 2,093 | 836 | -60% | 0 | 0 | — |
case-19 | pass→pass | 5,089 | 6,558 | +29% | 1 | 1 | 0% | 845 | 1,490 | +76% | 0 | 0 | — |
case-20 | pass→pass | 7,517 | 2,793 | -63% | 1 | 1 | 0% | 1,303 | 851 | -35% | 0 | 0 | — |
case-21 | fail→pass | 4,952 | 1,581 | -68% | 1 | 1 | 0% | 837 | 597 | -29% | 0 | 0 | — |
case-22 | fail→fail | 16,589 | 18,507 | +12% | 1 | 1 | 0% | 3,308 | 3,643 | +10% | 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 17 counted toward the lift figure. The other 5 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 17 comparable cases. 1 case got worse with the skill loaded, and it is 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.