Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Phase 4 Turn 2 of disease-market-sizing-orchestration (A6' upgrade). Audits compose output via Cite-or-Block strict — every anchor must reverse-resolve to an actual file in sources/, every fact claim must have an anchor. Returns verdict (ok/blocked) + violations list. Use AFTER cite-bound-content-generator (Turn 1), BEFORE report-bundle-builder (phase 5).
.claude/skills/ethanyoq-content-verification-layer/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 689% | 0% |
| case-02 | ✗→✓ | ▲ Improved | -45% | 0% |
| case-03 | ✗→✓ | ▲ Improved | -62% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -38% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -50% | 0% |
enforce_citations_in_text(html) — flag fact-bearing claims without anchorsverify_section(section_html, sources_dir) -> Verdictverify_full_report(html, sources_dir) -> Verdictenforce_citations_in_text(html) -> list[Violation]Tests grep skill source for hardcoded drug names / disease names — CI blocks on hit.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 4,686 | 28,903 | +517% | 1 | 1 | 0% | 806 | 6,362 | +689% | 0 | 0 | — |
case-02 | fail→pass | 11,638 | 5,438 | -53% | 1 | 1 | 0% | 2,480 | 1,359 | -45% | 0 | 0 | — |
case-03 | fail→pass | 10,308 | 4,032 | -61% | 1 | 1 | 0% | 2,235 | 853 | -62% | 0 | 0 | — |
case-04 | fail→pass | 9,746 | 4,619 | -53% | 1 | 1 | 0% | 2,008 | 1,236 | -38% | 0 | 0 | — |
case-05 | pass→pass | 11,496 | 4,059 | -65% | 1 | 1 | 0% | 2,290 | 1,030 | -55% | 0 | 0 | — |
case-06 | fail→pass | 5,605 | 1,773 | -68% | 1 | 1 | 0% | 1,037 | 514 | -50% | 0 | 0 | — |
case-07 | fail→pass | 8,684 | 2,417 | -72% | 1 | 1 | 0% | 1,605 | 716 | -55% | 0 | 0 | — |
case-08 | pass→pass | 6,759 | 5,660 | -16% | 1 | 1 | 0% | 1,475 | 1,597 | +8% | 0 | 0 | — |
case-09 | pass→pass | 4,817 | 1,301 | -73% | 1 | 1 | 0% | 894 | 469 | -48% | 0 | 0 | — |
case-10 | pass→pass | 3,866 | 3,362 | -13% | 1 | 1 | 0% | 794 | 915 | +15% | 0 | 0 | — |
case-11 | pass→pass | 7,503 | 2,517 | -66% | 1 | 1 | 0% | 1,351 | 727 | -46% | 0 | 0 | — |
case-12 | fail→pass | 9,652 | 3,030 | -69% | 1 | 1 | 0% | 1,934 | 865 | -55% | 0 | 0 | — |
case-13 | fail→pass | 8,922 | 1,077 | -88% | 1 | 1 | 0% | 1,721 | 419 | -76% | 0 | 0 | — |
case-14 | pass→pass | 6,478 | 3,315 | -49% | 1 | 1 | 0% | 1,209 | 877 | -27% | 0 | 0 | — |
case-15 | pass→pass | 7,274 | 1,673 | -77% | 1 | 1 | 0% | 1,496 | 547 | -63% | 0 | 0 | — |
case-16 | fail→pass | 13,209 | 7,149 | -46% | 1 | 1 | 0% | 2,555 | 1,703 | -33% | 0 | 0 | — |
case-21 | pass→pass | 2,374 | 2,978 | +25% | 1 | 1 | 0% | 532 | 689 | +30% | 0 | 0 | — |
case-17 | fail→pass | 32,003 | 2,951 | -91% | 1 | 1 | 0% | 1,916 | 788 | -59% | 0 | 0 | — |
case-18 | pass→pass | 5,987 | 1,470 | -75% | 1 | 1 | 0% | 1,205 | 523 | -57% | 0 | 0 | — |
case-19 | fail→pass | 19,489 | 1,298 | -93% | 1 | 1 | 0% | 4,076 | 426 | -90% | 0 | 0 | — |
case-20 | fail→pass | 10,112 | 1,813 | -82% | 1 | 1 | 0% | 1,776 | 537 | -70% | 0 | 0 | — |
case-22 | pass→pass | 2,618 | 2,978 | +14% | 1 | 1 | 0% | 558 | 787 | +41% | 0 | 0 | — |
case-23 | pass→pass | 2,702 | 2,311 | -14% | 1 | 1 | 0% | 579 | 702 | +21% | 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. The headline lift of +52 percentage points is the difference between those two pass rates over the 23 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.