Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use when `evidence-review` has screened includes and needs a schema-aligned extraction table. **Trigger**: extraction form, extraction table, data extraction, 信息提取, 提取表. **Use when**: `evidence-review` 在 screening 后进入 extraction（C4），需要把纳入论文按字段落到 CSV 以支持后续 synthesis。
.claude/skills/willoscar-extraction-form/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-05 | ✗→✓ | ▲ Improved | -60% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -35% | 0% |
| case-09 | ✗→✓ | ▲ Improved | -66% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -58% | 0% |
| case-13 | ✗→✓ | ▲ Improved | -36% | 0% |
Transforms screened include rows plus protocol schema into the analysis table used by evidence-review.
Required:
papers/screening_log.csvoutput/PROTOCOL.mdOptional:
papers/paper_notes.jsonlpapers/extraction_table.csvThe table must:
paper_id, title, year, url)notes, not in schema columnsscripts/run.py should:
include rowsoutput/PROTOCOL.md| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 5,169 | 4,537 | -12% | 1 | 1 | 0% | 714 | 389 | -46% | 0 | 0 | — |
case-02 | fail→fail | 13,510 | 4,823 | -64% | 1 | 1 | 0% | 2,345 | 438 | -81% | 0 | 0 | — |
case-03 | fail→fail | 7,941 | 4,114 | -48% | 1 | 1 | 0% | 1,174 | 330 | -72% | 0 | 0 | — |
case-04 | fail→fail | 14,908 | 19,781 | +33% | 1 | 1 | 0% | 2,660 | 3,914 | +47% | 0 | 0 | — |
case-05 | fail→pass | 11,699 | 3,396 | -71% | 1 | 1 | 0% | 1,861 | 750 | -60% | 0 | 0 | — |
case-06 | fail→pass | 6,840 | 2,847 | -58% | 1 | 1 | 0% | 1,012 | 653 | -35% | 0 | 0 | — |
case-07 | pass→pass | 11,340 | 3,374 | -70% | 1 | 1 | 0% | 1,699 | 809 | -52% | 0 | 0 | — |
case-08 | pass→pass | 12,883 | 5,199 | -60% | 1 | 1 | 0% | 1,704 | 958 | -44% | 0 | 0 | — |
case-09 | fail→pass | 9,782 | 2,190 | -78% | 1 | 1 | 0% | 1,535 | 527 | -66% | 0 | 0 | — |
case-10 | fail→pass | 12,989 | 3,735 | -71% | 1 | 1 | 0% | 2,053 | 858 | -58% | 0 | 0 | — |
case-11 | fail→fail | 13,247 | 4,213 | -68% | 1 | 1 | 0% | 1,987 | 859 | -57% | 0 | 0 | — |
case-12 | pass→pass | 4,595 | 1,439 | -69% | 1 | 1 | 0% | 690 | 397 | -42% | 0 | 0 | — |
case-13 | fail→pass | 5,318 | 1,850 | -65% | 1 | 1 | 0% | 807 | 516 | -36% | 0 | 0 | — |
case-14 | pass→pass | 10,892 | 2,778 | -74% | 1 | 1 | 0% | 1,640 | 635 | -61% | 0 | 0 | — |
case-15 | fail→fail | 10,138 | 7,081 | -30% | 1 | 1 | 0% | 1,594 | 1,438 | -10% | 0 | 0 | — |
case-16 | fail→pass | 8,766 | 2,613 | -70% | 1 | 1 | 0% | 1,294 | 626 | -52% | 0 | 0 | — |
case-17 | pass→pass | 11,848 | 4,227 | -64% | 1 | 1 | 0% | 1,770 | 944 | -47% | 0 | 0 | — |
case-18 | pass→pass | 11,234 | 3,099 | -72% | 1 | 1 | 0% | 1,620 | 717 | -56% | 0 | 0 | — |
case-19 | pass→pass | 12,145 | 3,215 | -74% | 1 | 1 | 0% | 1,829 | 691 | -62% | 0 | 0 | — |
case-20 | fail→pass | 9,164 | 3,242 | -65% | 1 | 1 | 0% | 1,400 | 774 | -45% | 0 | 0 | — |
case-21 | pass→pass | 2,219 | 1,883 | -15% | 1 | 1 | 0% | 307 | 488 | +59% | 0 | 0 | — |
case-22 | pass→pass | 11,998 | 3,782 | -68% | 1 | 1 | 0% | 1,815 | 838 | -54% | 0 | 0 | — |
case-23 | fail→pass | 14,939 | 8,886 | -41% | 1 | 1 | 0% | 2,297 | 1,646 | -28% | 0 | 0 | — |
case-24 | fail→pass | 6,051 | 6,997 | +16% | 1 | 1 | 0% | 924 | 1,305 | +41% | 0 | 0 | — |
case-25 | fail→fail | 3,323 | 11,255 | +239% | 1 | 1 | 0% | 513 | 1,635 | +219% | 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. 25 cases were attempted, and 22 counted toward the lift figure. The other 3 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 +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.