Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Combinatoric grammar that deterministically generates whole runnable child projects (src/, tests/, scripts/, manuscript/) from a seed — recompute-based provenance verification and a falsifiable honesty manifest against green-by-construction test theater.
.claude/skills/docxology-template-autopoiesis/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 26% | 0% |
| case-02 | ✗→✓ | ▲ Improved | -22% | 0% |
| case-03 | ✗→✓ | ▲ Improved | -23% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -10% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 45% | 0% |
Project-scoped skill for the in-repo exemplar at projects/templates/template_autopoiesis/. Load this when working inside the project.
template_autopoiesis exemplar — running the grammar,materializing child projects, or editing source.
recompute-verified provenance still hold after a grammar change.
bash# From the repository root uv run pytest projects/templates/template_autopoiesis/tests --cov=projects/templates/template_autopoiesis/src --cov-fail-under=90 # Materialize a child project from the grammar (writes a full runnable tree) uv run python projects/templates/template_autopoiesis/scripts/autopoiesis.py materialize --out-root /tmp/children # Realize a full archetype end-to-end (spec → child → seal) uv run python projects/templates/template_autopoiesis/scripts/realize_child_full.py # Seal a materialized child with QR provenance uv run python projects/templates/template_autopoiesis/scripts/seal_child.py
src/ (grammar.py,expand.py, materialize.py, realize.py, sealing.py, verify.py), not in scripts/.
exemplar is that a child's provenance is re-derived from disk at verify time, not read from a cached manifest.
output/; regenerate from thegrammar in manuscript/config.yaml.
scripts/01_generate_manuscript_assets.pyrequires output/data/coverage_full.json and writes output/figures/figure_registry.json only after all four referenced PNGs exist.
working directory unless the child AGENTS.md states otherwise.
AGENTS.mdREADME.mdSYNTAX.mdSTANDALONE.mdTODO.mdprojects/AGENTS.md| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 8,340 | 4,740 | -43% | 1 | 1 | 0% | 1,313 | 1,648 | +26% | 0 | 0 | — |
case-02 | fail→pass | 13,992 | 5,814 | -58% | 1 | 1 | 0% | 2,285 | 1,772 | -22% | 0 | 0 | — |
case-03 | fail→pass | 15,127 | 6,219 | -59% | 1 | 1 | 0% | 2,267 | 1,752 | -23% | 0 | 0 | — |
case-04 | fail→pass | 10,854 | 3,993 | -63% | 1 | 1 | 0% | 1,460 | 1,312 | -10% | 0 | 0 | — |
case-05 | pass→pass | 5,869 | 5,551 | -5% | 1 | 1 | 0% | 967 | 1,066 | +10% | 0 | 0 | — |
case-06 | pass→pass | 12,458 | 8,310 | -33% | 1 | 1 | 0% | 2,000 | 2,084 | +4% | 0 | 0 | — |
case-07 | fail→pass | 8,176 | 2,272 | -72% | 1 | 1 | 0% | 688 | 999 | +45% | 0 | 0 | — |
case-08 | fail→pass | 6,895 | 4,943 | -28% | 1 | 1 | 0% | 1,198 | 1,470 | +23% | 0 | 0 | — |
case-09 | pass→pass | 14,777 | 5,849 | -60% | 1 | 1 | 0% | 2,630 | 1,725 | -34% | 0 | 0 | — |
case-10 | pass→pass | 11,580 | 2,590 | -78% | 1 | 1 | 0% | 1,986 | 1,084 | -45% | 0 | 0 | — |
case-11 | pass→pass | 13,975 | 3,223 | -77% | 1 | 1 | 0% | 2,255 | 1,106 | -51% | 0 | 0 | — |
case-12 | fail→pass | 13,938 | 5,104 | -63% | 1 | 1 | 0% | 2,285 | 1,459 | -36% | 0 | 0 | — |
case-13 | pass→pass | 7,142 | 3,476 | -51% | 1 | 1 | 0% | 1,319 | 1,282 | -3% | 0 | 0 | — |
case-14 | fail→pass | 11,946 | 1,749 | -85% | 1 | 1 | 0% | 2,116 | 926 | -56% | 0 | 0 | — |
case-15 | fail→pass | 8,007 | 3,613 | -55% | 1 | 1 | 0% | 1,421 | 1,396 | -2% | 0 | 0 | — |
case-16 | pass→pass | 4,357 | 3,003 | -31% | 1 | 1 | 0% | 715 | 1,072 | +50% | 0 | 0 | — |
case-17 | fail→pass | 12,872 | 2,018 | -84% | 1 | 1 | 0% | 2,140 | 988 | -54% | 0 | 0 | — |
case-18 | fail→pass | 29,447 | 2,117 | -93% | 1 | 1 | 0% | 1,429 | 895 | -37% | 0 | 0 | — |
case-19 | fail→pass | 12,559 | 2,815 | -78% | 1 | 1 | 0% | 1,937 | 1,139 | -41% | 0 | 0 | — |
case-20 | pass→pass | 8,902 | 2,932 | -67% | 1 | 1 | 0% | 1,562 | 1,155 | -26% | 0 | 0 | — |
case-21 | pass→pass | 9,375 | 3,476 | -63% | 1 | 1 | 0% | 1,648 | 1,223 | -26% | 0 | 0 | — |
case-22 | pass→pass | 10,174 | 3,793 | -63% | 1 | 1 | 0% | 1,488 | 1,234 | -17% | 0 | 0 | — |
case-23 | pass→pass | 7,435 | 2,183 | -71% | 1 | 1 | 0% | 1,231 | 945 | -23% | 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, and 22 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 +52 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.
Other measured skills in the registry, with their headline benchmark lift.