Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Paper full text access via alphaxiv answer_pdf_queries or get_paper_content(fullText=true). Import of literature-engine/paper-research skill. Raw extracted text for precise claims.
.claude/skills/yogsoth-ai-stress-test-paper-research/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-21 | ✓→✗ | ▼ Worse | -16% | 0% |
| case-20 | ✓→✓ | ＝ Same ✓ | -4% | 0% |
| case-22 | ✓→✓ | ＝ Same ✓ | -15% | 0% |
| case-05 | ✗→✗ | ＝ Same ✗ | 128% | 0% |
| case-04 | ✗→✗ | ＝ Same ✗ | -54% | 0% |
Paper full text access — raw extracted text for precise claims and exact data.
Import — strictly follow literature-engine/paper-research skill protocol.
Returns full paper text. Use for exact quotes, detailed methodology, specific numbers, or any claim requiring precision beyond AI summary.
Quantity target is set by the calling strategy's budget table. This SOP executes one unit = one answer_pdf_queries or get_paper_content(fullText=true) call.
literature-engine repo → skills/paper-research/SKILL.md
<!-- 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-05 | fail→fail | 20,257 | 39,770 | +96% | 1 | 1 | 0% | 2,813 | 6,410 | +128% | 0 | 0 | — |
case-04 | fail→fail | 23,223 | 11,817 | -49% | 1 | 1 | 0% | 3,131 | 1,442 | -54% | 0 | 0 | — |
case-01 | fail→fail | 10,625 | 16,080 | +51% | 1 | 1 | 0% | 1,392 | 515 | -63% | 0 | 0 | — |
case-02 | fail→fail | 7,279 | 21,995 | +202% | 1 | 1 | 0% | 1,207 | 1,133 | -6% | 0 | 0 | — |
case-03 | fail→fail | 21,795 | 22,842 | +5% | 1 | 1 | 0% | 2,816 | 3,855 | +37% | 0 | 0 | — |
case-06 | fail→fail | 45,715 | 35,112 | -23% | 1 | 1 | 0% | 8,230 | 6,461 | -21% | 0 | 0 | — |
case-07 | fail→fail | 43,340 | 9,669 | -78% | 1 | 1 | 0% | 6,741 | 810 | -88% | 0 | 0 | — |
case-08 | fail→fail | 23,852 | 9,543 | -60% | 1 | 1 | 0% | 3,018 | 469 | -84% | 0 | 0 | — |
case-09 | fail→fail | 29,536 | 9,602 | -67% | 1 | 1 | 0% | 4,394 | 996 | -77% | 0 | 0 | — |
case-10 | fail→fail | 34,285 | 33,947 | -1% | 1 | 1 | 0% | 4,920 | 5,332 | +8% | 0 | 0 | — |
case-11 | fail→fail | 30,067 | 14,267 | -53% | 1 | 1 | 0% | 4,655 | 855 | -82% | 0 | 0 | — |
case-12 | fail→fail | 47,206 | 10,564 | -78% | 1 | 1 | 0% | 8,224 | 651 | -92% | 0 | 0 | — |
case-13 | fail→fail | 18,581 | 12,474 | -33% | 1 | 1 | 0% | 2,239 | 634 | -72% | 0 | 0 | — |
case-14 | fail→fail | 20,571 | 12,941 | -37% | 1 | 1 | 0% | 3,488 | 852 | -76% | 0 | 0 | — |
case-15 | fail→fail | 42,269 | 32,813 | -22% | 1 | 1 | 0% | 7,886 | 3,726 | -53% | 0 | 0 | — |
case-16 | fail→fail | 13,949 | 6,685 | -52% | 1 | 1 | 0% | 1,491 | 478 | -68% | 0 | 0 | — |
case-17 | fail→fail | 9,887 | 23,173 | +134% | 1 | 1 | 0% | 781 | 3,471 | +344% | 0 | 0 | — |
case-18 | fail→fail | 22,542 | 26,834 | +19% | 1 | 1 | 0% | 3,207 | 667 | -79% | 0 | 0 | — |
case-19 | fail→fail | 11,423 | 10,130 | -11% | 1 | 1 | 0% | 976 | 436 | -55% | 0 | 0 | — |
case-20 | pass→pass | 13,382 | 11,804 | -12% | 1 | 1 | 0% | 1,620 | 1,556 | -4% | 0 | 0 | — |
case-21 | pass→fail | 26,357 | 28,624 | +9% | 1 | 1 | 0% | 5,903 | 4,962 | -16% | 0 | 0 | — |
case-22 | pass→pass | 17,989 | 13,947 | -22% | 1 | 1 | 0% | 2,265 | 1,928 | -15% | 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 12 counted toward the lift figure. The other 10 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 -5 percentage points is the difference between those two pass rates over the 12 comparable cases. 3 cases got worse with the skill loaded, and they are included in that figure.
Other measured skills in the registry, with their headline benchmark lift.