Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Full-depth paper reading with raw text extraction. Import of literature-engine/literature-research skill. Must read fullText (true) — equations, hyperparameters, specific claims extracted.
.claude/skills/yogsoth-ai-knowledge-acquisition-paper-research/SKILL.md| Model | Eval pass | Runs |
|---|---|---|
| gemini-3.6-flash | 100% | 18 |
| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-08 | ✗→✓ | ▲ Improved | -63% | 0% |
| case-17 | ✗→✓ | ▲ Improved | -88% | 0% |
| case-22 | ✗→✓ | ▲ Improved | -31% | 0% |
| case-23 | ✗→✓ | ▲ Improved | -65% | 0% |
| case-04 | ✓→✗ | ▼ Worse | -75% | 0% |
Full-depth paper reading with raw text extraction.
Import — strictly follow literature-engine/literature-research skill protocol.
Must read with fullText: true — extract equations, hyperparameters, specific claims, and implementation details. This is the deepest reading level.
Quantity target is set by the calling strategy's budget table. This SOP executes one unit = one paper read at full depth.
literature-engine repo → skills/literature-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-01 | pass→pass | 46,836 | 35,731 | -24% | 1 | 1 | 0% | 311 | 563 | +81% | 0 | 0 | — |
case-02 | pass→pass | 47,063 | 45,855 | -3% | 1 | 1 | 0% | 3,237 | 2,834 | -12% | 0 | 0 | — |
case-03 | pass→pass | 29,438 | 10,785 | -63% | 1 | 1 | 0% | 1,030 | 757 | -27% | 0 | 0 | — |
case-04 | pass→fail | 19,629 | 28,302 | +44% | 1 | 1 | 0% | 2,306 | 588 | -75% | 0 | 0 | — |
case-05 | pass→pass | 16,284 | 15,555 | -4% | 1 | 1 | 0% | 892 | 904 | +1% | 0 | 0 | — |
case-06 | pass→pass | 17,823 | 14,455 | -19% | 1 | 1 | 0% | 2,213 | 1,354 | -39% | 0 | 0 | — |
case-07 | pass→pass | 15,897 | 13,348 | -16% | 1 | 1 | 0% | 1,470 | 1,433 | -3% | 0 | 0 | — |
case-08 | fail→pass | 13,939 | 8,321 | -40% | 1 | 1 | 0% | 1,116 | 414 | -63% | 0 | 0 | — |
case-09 | fail→fail | 16,230 | 15,153 | -7% | 1 | 1 | 0% | 1,538 | 963 | -37% | 0 | 0 | — |
case-10 | pass→pass | 23,850 | 8,019 | -66% | 1 | 1 | 0% | 2,240 | 732 | -67% | 0 | 0 | — |
case-11 | pass→pass | 14,334 | 9,514 | -34% | 1 | 1 | 0% | 1,258 | 901 | -28% | 0 | 0 | — |
case-12 | pass→pass | 20,607 | 11,430 | -45% | 1 | 1 | 0% | 2,504 | 1,118 | -55% | 0 | 0 | — |
case-13 | pass→fail | 8,267 | 2,887 | -65% | 1 | 1 | 0% | 479 | 461 | -4% | 0 | 0 | — |
case-14 | pass→pass | 10,289 | 16,186 | +57% | 1 | 1 | 0% | 1,600 | 1,330 | -17% | 0 | 0 | — |
case-15 | pass→pass | 12,252 | 9,147 | -25% | 1 | 1 | 0% | 2,009 | 619 | -69% | 0 | 0 | — |
case-24 | fail→fail | 12,876 | 3,076 | -76% | 1 | 1 | 0% | 2,227 | 560 | -75% | 0 | 0 | — |
case-16 | fail→fail | 7,619 | 1,735 | -77% | 1 | 1 | 0% | 1,174 | 484 | -59% | 0 | 0 | — |
case-17 | fail→pass | 22,984 | 2,493 | -89% | 1 | 1 | 0% | 3,025 | 354 | -88% | 0 | 0 | — |
case-18 | pass→pass | 23,683 | 10,539 | -55% | 1 | 1 | 0% | 2,687 | 1,072 | -60% | 0 | 0 | — |
case-19 | pass→pass | 11,314 | 16,418 | +45% | 1 | 1 | 0% | 1,612 | 842 | -48% | 0 | 0 | — |
case-20 | pass→pass | 22,116 | 15,940 | -28% | 1 | 1 | 0% | 2,106 | 1,942 | -8% | 0 | 0 | — |
case-21 | pass→pass | 10,503 | 3,312 | -68% | 1 | 1 | 0% | 1,246 | 825 | -34% | 0 | 0 | — |
case-22 | fail→pass | 52,347 | 23,592 | -55% | 1 | 1 | 0% | 8,213 | 5,694 | -31% | 0 | 0 | — |
case-23 | fail→pass | 17,896 | 28,034 | +57% | 1 | 1 | 0% | 2,188 | 772 | -65% | 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. 24 cases were attempted. The headline lift of +8 percentage points is the difference between those two pass rates over the 24 comparable cases. 2 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.