Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Derived documentation generators for the template research framework. Scripts that write to docs/_generated/ and update in-place doc blocks. Currently at scripts/ root; this directory is the planned migration target.
.claude/skills/docxology-template-docgen/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | -70% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -2% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -18% | 0% |
| case-09 | ✗→✓ | ▲ Improved | -71% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -68% | 0% |
Documentation generation scripts under scripts/docgen/.
Load this skill when you need to regenerate derived documentation:
docs/_generated/__all__bashuv run python scripts/docgen/stage_table.py uv run python scripts/docgen/api_reference.py --write uv run python scripts/docgen/active_projects.py --write uv run python scripts/docgen/architecture_overview.py --write uv run python scripts/docgen/counts.py --write
infrastructure/core/pipeline/pipeline.yaml.api_reference.py --check is run in CI — do not break exports.docs/_generated/ must not be hand-edited.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-06 | fail→fail | 13,278 | 3,201 | -76% | 1 | 1 | 0% | 1,904 | 579 | -70% | 0 | 0 | — |
case-01 | pass→pass | 7,172 | 33,194 | +363% | 1 | 1 | 0% | 1,137 | 772 | -32% | 0 | 0 | — |
case-02 | fail→pass | 10,097 | 1,825 | -82% | 1 | 1 | 0% | 1,642 | 494 | -70% | 0 | 0 | — |
case-03 | fail→fail | 7,811 | 5,079 | -35% | 1 | 1 | 0% | 1,161 | 997 | -14% | 0 | 0 | — |
case-04 | fail→fail | 18,270 | 3,358 | -82% | 1 | 1 | 0% | 3,038 | 839 | -72% | 0 | 0 | — |
case-05 | fail→pass | 3,796 | 2,372 | -38% | 1 | 1 | 0% | 528 | 517 | -2% | 0 | 0 | — |
case-07 | fail→fail | 5,771 | 3,447 | -40% | 1 | 1 | 0% | 831 | 647 | -22% | 0 | 0 | — |
case-08 | fail→pass | 6,334 | 3,463 | -45% | 1 | 1 | 0% | 909 | 742 | -18% | 0 | 0 | — |
case-09 | fail→pass | 9,334 | 1,610 | -83% | 1 | 1 | 0% | 1,347 | 394 | -71% | 0 | 0 | — |
case-10 | fail→pass | 8,879 | 1,590 | -82% | 1 | 1 | 0% | 1,393 | 440 | -68% | 0 | 0 | — |
case-11 | fail→pass | 13,483 | 2,387 | -82% | 1 | 1 | 0% | 1,979 | 578 | -71% | 0 | 0 | — |
case-12 | fail→pass | 11,478 | 1,777 | -85% | 1 | 1 | 0% | 1,685 | 480 | -72% | 0 | 0 | — |
case-13 | fail→pass | 7,075 | 2,144 | -70% | 1 | 1 | 0% | 933 | 520 | -44% | 0 | 0 | — |
case-14 | pass→pass | 8,826 | 1,832 | -79% | 1 | 1 | 0% | 1,512 | 494 | -67% | 0 | 0 | — |
case-20 | pass→pass | 10,718 | 2,713 | -75% | 1 | 1 | 0% | 1,519 | 611 | -60% | 0 | 0 | — |
case-15 | fail→pass | 11,224 | 3,420 | -70% | 1 | 1 | 0% | 1,640 | 777 | -53% | 0 | 0 | — |
case-16 | fail→pass | 2,276 | 1,492 | -34% | 1 | 1 | 0% | 302 | 382 | +26% | 0 | 0 | — |
case-17 | fail→pass | 5,548 | 1,442 | -74% | 1 | 1 | 0% | 772 | 362 | -53% | 0 | 0 | — |
case-18 | fail→pass | 14,470 | 1,759 | -88% | 1 | 1 | 0% | 2,106 | 472 | -78% | 0 | 0 | — |
case-19 | pass→pass | 7,822 | 3,579 | -54% | 1 | 1 | 0% | 1,279 | 769 | -40% | 0 | 0 | — |
case-21 | pass→pass | 14,948 | 2,492 | -83% | 1 | 1 | 0% | 2,435 | 566 | -77% | 0 | 0 | — |
case-22 | fail→pass | 13,113 | 2,254 | -83% | 1 | 1 | 0% | 1,890 | 607 | -68% | 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. The headline lift of +59 percentage points is the difference between those two pass rates over the 22 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.
| Model | Method | Date | Lift |
|---|---|---|---|
| gemini-3.6-flash | verified | 8/4/2026 | +70% |
Other measured skills in the registry, with their headline benchmark lift.