Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Mid-depth paper analysis via AI-generated summaries. Import of literature-engine/literature-search skill. Reads AI summary — sufficient for methodology understanding but not for quoting specific results.
.claude/skills/yogsoth-ai-creative-ideation-paper-search/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-03 | ✗→✓ | ▲ Improved | -47% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -13% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -64% | 0% |
| case-06 | ✓→✗ | ▼ Worse | -83% | 0% |
| case-09 | ✓→✗ | ▼ Worse | 42% | 0% |
Mid-depth paper analysis via AI-generated summaries.
Import — strictly follow literature-engine/literature-search skill protocol.
AI summary level — sufficient for understanding methodology and approach, but not for quoting specific numerical results or exact claims. Use paper-research for that.
Quantity target is set by the calling strategy's budget table. This SOP executes one unit = one paper read at AI summary level.
literature-engine repo → skills/literature-search/SKILL.md
<!-- BEGIN available-tables (generated) -->
Optional, no fixed order; the final leaf is always a sop.
| SOP | When to use | | --- | --- | | literature-search | Medium-depth literature search — read AI-summarized reports for every paper analyzed |
<!-- 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 | 5,454 | 5,217 | -4% | 1 | 1 | 0% | 247 | 437 | +77% | 0 | 0 | — |
case-02 | fail→fail | 19,451 | 22,349 | +15% | 1 | 1 | 0% | 3,410 | 4,058 | +19% | 0 | 0 | — |
case-03 | fail→pass | 12,289 | 6,774 | -45% | 1 | 1 | 0% | 2,470 | 1,312 | -47% | 0 | 0 | — |
case-04 | fail→pass | 18,418 | 15,232 | -17% | 1 | 1 | 0% | 3,447 | 2,984 | -13% | 0 | 0 | — |
case-05 | fail→pass | 20,640 | 7,247 | -65% | 1 | 1 | 0% | 3,934 | 1,412 | -64% | 0 | 0 | — |
case-06 | pass→fail | 22,113 | 7,566 | -66% | 1 | 1 | 0% | 3,665 | 614 | -83% | 0 | 0 | — |
case-07 | pass→pass | 23,123 | 31,352 | +36% | 1 | 1 | 0% | 3,605 | 5,032 | +40% | 0 | 0 | — |
case-08 | pass→pass | 19,707 | 21,272 | +8% | 1 | 1 | 0% | 3,283 | 3,772 | +15% | 0 | 0 | — |
case-09 | pass→fail | 22,226 | 30,249 | +36% | 1 | 1 | 0% | 3,568 | 5,073 | +42% | 0 | 0 | — |
case-10 | pass→pass | 11,343 | 2,437 | -79% | 1 | 1 | 0% | 1,946 | 570 | -71% | 0 | 0 | — |
case-11 | fail→fail | 19,364 | 27,630 | +43% | 1 | 1 | 0% | 3,142 | 4,203 | +34% | 0 | 0 | — |
case-12 | pass→pass | 22,075 | 24,553 | +11% | 1 | 1 | 0% | 3,239 | 4,066 | +26% | 0 | 0 | — |
case-13 | pass→pass | 22,605 | 22,932 | +1% | 1 | 1 | 0% | 3,807 | 3,916 | +3% | 0 | 0 | — |
case-14 | pass→pass | 21,379 | 22,993 | +8% | 1 | 1 | 0% | 3,402 | 3,734 | +10% | 0 | 0 | — |
case-15 | fail→fail | 20,027 | 27,443 | +37% | 1 | 1 | 0% | 3,137 | 4,040 | +29% | 0 | 0 | — |
case-16 | pass→fail | 20,216 | 44,962 | +122% | 1 | 1 | 0% | 3,105 | 4,008 | +29% | 0 | 0 | — |
case-17 | fail→fail | 22,628 | 4,603 | -80% | 1 | 1 | 0% | 3,576 | 397 | -89% | 0 | 0 | — |
case-18 | fail→fail | 22,294 | 43,057 | +93% | 1 | 1 | 0% | 3,523 | 679 | -81% | 0 | 0 | — |
case-19 | fail→fail | 18,138 | 22,776 | +26% | 1 | 1 | 0% | 2,756 | 3,754 | +36% | 0 | 0 | — |
case-20 | pass→pass | 17,808 | 27,389 | +54% | 1 | 1 | 0% | 2,859 | 4,735 | +66% | 0 | 0 | — |
case-21 | pass→fail | 17,984 | 6,302 | -65% | 1 | 1 | 0% | 3,008 | 480 | -84% | 0 | 0 | — |
case-22 | pass→pass | 20,970 | 15,637 | -25% | 1 | 1 | 0% | 3,487 | 2,724 | -22% | 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 -5 percentage points is the difference between those two pass rates over the 17 comparable cases. 6 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.
Other measured skills in the registry, with their headline benchmark lift.