Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →End-to-end publication-readiness audit for public template exemplars — tests, no-mocks, methods contracts, claims, evidence, figures, references, rendered outputs, and deterministic regeneration. USE WHEN preparing a template or project for release, DOI deposit, peer review, or a publication sign-off.
.claude/skills/docxology-template-publication-audit/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-10 | ✗→✓ | ▲ Improved | -44% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -48% | 0% |
| case-09 | ✗→✓ | ▲ Improved | -27% | 0% |
| case-12 | ✗→✓ | ▲ Improved | -45% | 0% |
| case-13 | ✗→✓ | ▲ Improved | -10% | 0% |
Use the deterministic infrastructure gate first:
bashuv run python -m infrastructure.validation.cli publication-audit \ --project templates/template_code_project --strict --rendered --format markdown
Then run the project’s tests, methods plan, prerender gate, render, output validation, no-mock inventory, and clean double-run comparison. Treat only stable validator failures as blocking. Preserve review_required findings for human assessment of scientific scope, unsupported generalization, prose, and visual quality.
Never hand-edit generated output. Correct the source producer or manuscript source and regenerate the affected artifacts.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-10 | fail→pass | 12,057 | 5,164 | -57% | 1 | 1 | 0% | 1,956 | 1,100 | -44% | 0 | 0 | — |
case-04 | pass→pass | 9,966 | 7,090 | -29% | 1 | 1 | 0% | 2,040 | 1,777 | -13% | 0 | 0 | — |
case-01 | fail→fail | 4,164 | 4,772 | +15% | 1 | 1 | 0% | 159 | 461 | +190% | 0 | 0 | — |
case-02 | fail→fail | 7,949 | 4,234 | -47% | 1 | 1 | 0% | 216 | 435 | +101% | 0 | 0 | — |
case-03 | fail→fail | 4,683 | 3,896 | -17% | 1 | 1 | 0% | 197 | 421 | +114% | 0 | 0 | — |
case-05 | pass→pass | 9,483 | 5,841 | -38% | 1 | 1 | 0% | 2,036 | 1,286 | -37% | 0 | 0 | — |
case-06 | pass→pass | 9,327 | 6,430 | -31% | 1 | 1 | 0% | 1,654 | 1,222 | -26% | 0 | 0 | — |
case-07 | fail→pass | 11,063 | 3,757 | -66% | 1 | 1 | 0% | 1,764 | 920 | -48% | 0 | 0 | — |
case-08 | pass→pass | 7,919 | 3,371 | -57% | 1 | 1 | 0% | 1,412 | 763 | -46% | 0 | 0 | — |
case-09 | fail→pass | 6,118 | 3,571 | -42% | 1 | 1 | 0% | 1,096 | 803 | -27% | 0 | 0 | — |
case-11 | fail→fail | 11,253 | 3,531 | -69% | 1 | 1 | 0% | 1,969 | 604 | -69% | 0 | 0 | — |
case-12 | fail→pass | 11,379 | 5,644 | -50% | 1 | 1 | 0% | 2,099 | 1,154 | -45% | 0 | 0 | — |
case-13 | fail→pass | 8,443 | 5,822 | -31% | 1 | 1 | 0% | 1,462 | 1,311 | -10% | 0 | 0 | — |
case-14 | fail→pass | 9,976 | 7,306 | -27% | 1 | 1 | 0% | 1,623 | 1,320 | -19% | 0 | 0 | — |
case-15 | fail→pass | 7,131 | 3,334 | -53% | 1 | 1 | 0% | 1,150 | 765 | -33% | 0 | 0 | — |
case-16 | fail→fail | 14,104 | 1,557 | -89% | 1 | 1 | 0% | 2,216 | 412 | -81% | 0 | 0 | — |
case-17 | pass→pass | 10,382 | 4,023 | -61% | 1 | 1 | 0% | 1,539 | 853 | -45% | 0 | 0 | — |
case-18 | fail→pass | 7,248 | 4,032 | -44% | 1 | 1 | 0% | 1,233 | 861 | -30% | 0 | 0 | — |
case-19 | pass→pass | 3,510 | 2,889 | -18% | 1 | 1 | 0% | 457 | 672 | +47% | 0 | 0 | — |
case-20 | pass→fail | 8,090 | 2,113 | -74% | 1 | 1 | 0% | 1,432 | 635 | -56% | 0 | 0 | — |
case-21 | fail→pass | 14,891 | 6,334 | -57% | 1 | 1 | 0% | 2,415 | 1,245 | -48% | 0 | 0 | — |
case-22 | fail→pass | 10,256 | 3,931 | -62% | 1 | 1 | 0% | 1,604 | 673 | -58% | 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 19 counted toward the lift figure. The other 3 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 +41 percentage points is the difference between those two pass rates over the 19 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.