Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Import SOP: paper full-text reading (from literature-engine skill)
.claude/skills/yogsoth-ai-experiment-execution-paper-research/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-09 | ✗→✓ | ▲ Improved | -62% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -69% | 0% |
| case-14 | ✗→✓ | ▲ Improved | -39% | 0% |
| case-22 | ✗→✓ | ▲ Improved | -81% | 0% |
| case-03 | ✓→✓ | ＝ Same ✓ | 275% | 0% |
Paper full-text reading — imported from literature-engine skill.
Import — invokes literature-engine skill's paper-research SOP.
Available to all campaigns for deep reading of paper full text (experimental details, statistical methods, implementation specifics, appendix data, etc.).
answer_pdf_queries, get_paper_content (fullText mode)<!-- BEGIN available-tables (generated) -->
Optional, no fixed order; the final leaf is always a sop.
| SOP | When to use | | --- | --- | | literature-research | Deep literature research — raw full text reading and targeted PDF queries for rigorous analysis |
<!-- END available-tables (generated) -->
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 11,757 | 5,774 | -51% | 1 | 1 | 0% | 1,861 | 454 | -76% | 0 | 0 | — |
case-02 | fail→fail | 27,146 | 4,762 | -82% | 1 | 1 | 0% | 4,624 | 456 | -90% | 0 | 0 | — |
case-03 | pass→pass | 3,758 | 11,923 | +217% | 1 | 1 | 0% | 588 | 2,206 | +275% | 0 | 0 | — |
case-04 | pass→pass | 17,819 | 16,833 | -6% | 1 | 1 | 0% | 3,998 | 4,227 | +6% | 0 | 0 | — |
case-05 | pass→pass | 21,059 | 12,452 | -41% | 1 | 1 | 0% | 4,357 | 2,803 | -36% | 0 | 0 | — |
case-06 | fail→fail | 9,862 | 5,047 | -49% | 1 | 1 | 0% | 1,778 | 489 | -72% | 0 | 0 | — |
case-07 | fail→fail | 15,645 | 5,806 | -63% | 1 | 1 | 0% | 2,715 | 518 | -81% | 0 | 0 | — |
case-08 | pass→pass | 16,496 | 3,287 | -80% | 1 | 1 | 0% | 2,678 | 677 | -75% | 0 | 0 | — |
case-09 | fail→pass | 7,677 | 2,180 | -72% | 1 | 1 | 0% | 1,180 | 445 | -62% | 0 | 0 | — |
case-10 | fail→pass | 10,538 | 1,912 | -82% | 1 | 1 | 0% | 1,676 | 519 | -69% | 0 | 0 | — |
case-11 | fail→fail | 23,652 | 4,560 | -81% | 1 | 1 | 0% | 3,852 | 399 | -90% | 0 | 0 | — |
case-12 | fail→fail | 23,623 | 4,269 | -82% | 1 | 1 | 0% | 3,984 | 411 | -90% | 0 | 0 | — |
case-13 | fail→fail | 7,949 | 5,516 | -31% | 1 | 1 | 0% | 1,334 | 492 | -63% | 0 | 0 | — |
case-14 | fail→pass | 7,793 | 4,081 | -48% | 1 | 1 | 0% | 1,400 | 853 | -39% | 0 | 0 | — |
case-15 | fail→fail | 18,279 | 5,637 | -69% | 1 | 1 | 0% | 3,002 | 541 | -82% | 0 | 0 | — |
case-21 | fail→fail | 13,506 | 5,037 | -63% | 1 | 1 | 0% | 2,643 | 448 | -83% | 0 | 0 | — |
case-16 | pass→pass | 14,995 | 2,019 | -87% | 1 | 1 | 0% | 2,521 | 491 | -81% | 0 | 0 | — |
case-17 | fail→fail | 19,597 | 5,317 | -73% | 1 | 1 | 0% | 3,545 | 551 | -84% | 0 | 0 | — |
case-18 | fail→fail | 15,562 | 4,218 | -73% | 1 | 1 | 0% | 2,836 | 342 | -88% | 0 | 0 | — |
case-19 | fail→fail | 14,542 | 5,222 | -64% | 1 | 1 | 0% | 2,779 | 404 | -85% | 0 | 0 | — |
case-20 | fail→fail | 12,356 | 5,730 | -54% | 1 | 1 | 0% | 2,108 | 543 | -74% | 0 | 0 | — |
case-22 | fail→pass | 10,076 | 1,382 | -86% | 1 | 1 | 0% | 1,696 | 328 | -81% | 0 | 0 | — |
case-23 | fail→fail | 18,912 | 5,282 | -72% | 1 | 1 | 0% | 3,518 | 404 | -89% | 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. 23 cases were attempted, and 9 counted toward the lift figure. The other 14 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 +17 percentage points is the difference between those two pass rates over the 9 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.