Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Template-native research intake, literature search, source verification, synthesis, fact-checking, and systematic-review planning. USE WHEN the user asks to research a topic, build a literature corpus, fact-check claims, prepare a PRISMA-style review, or clarify a research question before manuscript work.
.claude/skills/docxology-template-deep-research/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-06 | ✗→✓ | ▲ Improved | -40% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 3% | 0% |
| case-09 | ✗→✓ | ▲ Improved | -59% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -20% | 0% |
| case-11 | ✗→✓ | ▲ Improved | -41% | 0% |
Template-native research workflow. This skill routes to existing repository systems instead of running an autonomous external agent suite.
docs/_generated/active_projects.md.infrastructure.search.literature and project-local corpora; record query, backend, date, DOI/arXiv IDs, and failures.fabricated/mismatch records, surface unverifiable/unchecked honestly, and never invent unavailable bibliographic fields. This is the tier-0 anti-leakage step: flag the gap, do not hallucinate the citation.bashuv run python -m infrastructure.search.literature search "QUERY" --max-results 20 uv run python -m infrastructure.reference.citation validate projects/<project>/manuscript/references.bib uv run python -m infrastructure.reference.verification verify projects/<project>/manuscript/references.bib --live --json uv run python -m infrastructure.validation.cli evidence projects/<project> --fail-on-issues
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-19 | pass→pass | 15,807 | 13,002 | -18% | 1 | 1 | 0% | 2,425 | 3,102 | +28% | 0 | 0 | — |
case-20 | pass→pass | 13,794 | 14,396 | +4% | 1 | 1 | 0% | 2,649 | 3,345 | +26% | 0 | 0 | — |
case-01 | fail→fail | 34,175 | 35,621 | +4% | 1 | 1 | 0% | 6,189 | 6,861 | +11% | 0 | 0 | — |
case-02 | fail→fail | 5,601 | 6,774 | +21% | 1 | 1 | 0% | 944 | 1,116 | +18% | 0 | 0 | — |
case-03 | fail→fail | 29,036 | 74,770 | +158% | 1 | 1 | 0% | 5,123 | 4,350 | -15% | 0 | 0 | — |
case-04 | fail→fail | 33,363 | 9,339 | -72% | 1 | 1 | 0% | 6,186 | 2,219 | -64% | 0 | 0 | — |
case-05 | pass→pass | 11,977 | 7,244 | -40% | 1 | 1 | 0% | 2,068 | 1,944 | -6% | 0 | 0 | — |
case-06 | fail→pass | 9,622 | 2,681 | -72% | 1 | 1 | 0% | 1,852 | 1,105 | -40% | 0 | 0 | — |
case-07 | fail→pass | 12,812 | 8,169 | -36% | 1 | 1 | 0% | 2,057 | 2,109 | +3% | 0 | 0 | — |
case-08 | pass→pass | 11,073 | 1,947 | -82% | 1 | 1 | 0% | 1,732 | 953 | -45% | 0 | 0 | — |
case-09 | fail→pass | 11,346 | 1,933 | -83% | 1 | 1 | 0% | 2,299 | 945 | -59% | 0 | 0 | — |
case-10 | fail→pass | 6,251 | 1,673 | -73% | 1 | 1 | 0% | 1,223 | 976 | -20% | 0 | 0 | — |
case-11 | fail→pass | 9,509 | 2,119 | -78% | 1 | 1 | 0% | 1,795 | 1,053 | -41% | 0 | 0 | — |
case-12 | fail→pass | 19,849 | 1,943 | -90% | 1 | 1 | 0% | 2,987 | 923 | -69% | 0 | 0 | — |
case-13 | pass→pass | 10,962 | 5,744 | -48% | 1 | 1 | 0% | 1,525 | 1,512 | -1% | 0 | 0 | — |
case-14 | fail→pass | 10,329 | 4,004 | -61% | 1 | 1 | 0% | 1,480 | 1,325 | -10% | 0 | 0 | — |
case-15 | fail→pass | 11,513 | 6,047 | -47% | 1 | 1 | 0% | 1,780 | 1,703 | -4% | 0 | 0 | — |
case-16 | fail→pass | 12,510 | 2,901 | -77% | 1 | 1 | 0% | 2,261 | 1,117 | -51% | 0 | 0 | — |
case-17 | fail→pass | 7,515 | 2,837 | -62% | 1 | 1 | 0% | 1,088 | 1,139 | +5% | 0 | 0 | — |
case-18 | fail→pass | 13,925 | 4,594 | -67% | 1 | 1 | 0% | 2,231 | 1,271 | -43% | 0 | 0 | — |
case-21 | pass→pass | 5,987 | 5,283 | -12% | 1 | 1 | 0% | 1,165 | 1,540 | +32% | 0 | 0 | — |
case-22 | pass→pass | 10,838 | 2,653 | -76% | 1 | 1 | 0% | 1,881 | 1,068 | -43% | 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 21 counted toward the lift figure. The other 1 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 +50 percentage points is the difference between those two pass rates over the 21 comparable cases. 1 case got worse with the skill loaded, and it is 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.