Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Redacted release-report exemplar for classification ceilings, source protection, source-safe ledgers, reviewer gates, and mosaic-risk checks.
.claude/skills/docxology-template-redacted-report/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 59% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -52% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -50% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -56% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -60% | 0% |
Load this skill when working inside projects/templates/template_redacted_report/.
bashuv run pytest projects/templates/template_redacted_report/tests --cov=projects/templates/template_redacted_report/src --cov-fail-under=90 uv run python scripts/pipeline/stage_01_test.py --project templates/template_redacted_report --project-only uv run python scripts/pipeline/stage_02_analysis.py --project templates/template_redacted_report
redaction_audit.json and release_ledger.json as text-free publicevidence surfaces; the comprehensive narrative packet remains in memory.
01_generate_release_artifacts.py thin and the normal Stage 02 allowlistseparate from opt-in development PDF generation.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 12,184 | 18,669 | +53% | 1 | 1 | 0% | 2,053 | 3,267 | +59% | 0 | 0 | — |
case-02 | fail→fail | 21,765 | 3,168 | -85% | 1 | 1 | 0% | 3,742 | 659 | -82% | 0 | 0 | — |
case-03 | fail→fail | 17,896 | 5,658 | -68% | 1 | 1 | 0% | 2,531 | 662 | -74% | 0 | 0 | — |
case-04 | fail→pass | 10,593 | 2,496 | -76% | 1 | 1 | 0% | 1,520 | 728 | -52% | 0 | 0 | — |
case-05 | pass→pass | 11,316 | 5,359 | -53% | 1 | 1 | 0% | 1,840 | 1,185 | -36% | 0 | 0 | — |
case-06 | fail→pass | 11,327 | 4,516 | -60% | 1 | 1 | 0% | 2,026 | 1,004 | -50% | 0 | 0 | — |
case-07 | fail→pass | 7,637 | 1,681 | -78% | 1 | 1 | 0% | 1,388 | 604 | -56% | 0 | 0 | — |
case-08 | fail→pass | 6,995 | 1,472 | -79% | 1 | 1 | 0% | 1,328 | 531 | -60% | 0 | 0 | — |
case-09 | pass→pass | 16,671 | 12,807 | -23% | 1 | 1 | 0% | 2,831 | 2,615 | -8% | 0 | 0 | — |
case-10 | fail→pass | 8,940 | 3,052 | -66% | 1 | 1 | 0% | 1,520 | 690 | -55% | 0 | 0 | — |
case-11 | fail→pass | 9,534 | 2,011 | -79% | 1 | 1 | 0% | 1,796 | 590 | -67% | 0 | 0 | — |
case-12 | pass→fail | 16,708 | 29,927 | +79% | 1 | 1 | 0% | 1,619 | 1,205 | -26% | 0 | 0 | — |
case-13 | pass→pass | 5,158 | 1,918 | -63% | 1 | 1 | 0% | 1,091 | 664 | -39% | 0 | 0 | — |
case-14 | fail→pass | 8,370 | 2,410 | -71% | 1 | 1 | 0% | 1,471 | 645 | -56% | 0 | 0 | — |
case-15 | pass→pass | 16,744 | 7,707 | -54% | 1 | 1 | 0% | 2,166 | 1,532 | -29% | 0 | 0 | — |
case-16 | pass→pass | 12,862 | 4,816 | -63% | 1 | 1 | 0% | 1,997 | 995 | -50% | 0 | 0 | — |
case-17 | fail→pass | 7,630 | 2,475 | -68% | 1 | 1 | 0% | 1,274 | 718 | -44% | 0 | 0 | — |
case-18 | fail→pass | 9,219 | 4,117 | -55% | 1 | 1 | 0% | 1,525 | 927 | -39% | 0 | 0 | — |
case-19 | pass→pass | 11,710 | 14,264 | +22% | 1 | 1 | 0% | 1,891 | 2,402 | +27% | 0 | 0 | — |
case-20 | pass→pass | 10,154 | 8,778 | -14% | 1 | 1 | 0% | 1,913 | 1,729 | -10% | 0 | 0 | — |
case-21 | pass→pass | 4,910 | 4,252 | -13% | 1 | 1 | 0% | 671 | 1,098 | +64% | 0 | 0 | — |
case-22 | pass→pass | 10,003 | 8,827 | -12% | 1 | 1 | 0% | 1,718 | 1,795 | +4% | 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 20 counted toward the lift figure. The other 2 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 20 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.