Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Audit or author registry-driven manuscript cross-refs — labels.yaml, [[FIG:]], [[THMREF:]], [[VAR:]] tokens. USE WHEN fixing figure/equation/theorem numbering, orphan registry keys, hard-coded "Theorem 7.3" in prose, or [[MISSING:]] injection failures — even for Pandoc projects that also use a YAML registry.
.claude/skills/docxology-template-manuscript-cross-references/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-11 | ✗→✓ | ▲ Improved | -52% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 10% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -40% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -1% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 19% | 0% |
For Pandoc-crossref @fig: / [@key] exemplars, see docs/guides/manuscript-semantics.md. This skill is for YAML registry + inline tokens.
> Convention, not built-in infrastructure. refs/labels.yaml and [[FIG:]]/[[THMREF:]]/[[VAR:]] tokens are a project-authored convention, not a dedicated parser in infrastructure/: no current public exemplar under projects/templates/ uses this style (all ship Pandoc-crossref), and there is no labels.yaml-specific validator. Validation for a project that adopts this convention still runs through the generic infrastructure.validation.cli markdown/prerender checks below plus project-local test coverage the project author writes — token expansion is the project's own manuscript_variables.py-style substitution, not a shared registry engine.
projects/<name>/manuscript/ path or project name.refs/labels.yaml (figures, equations, sections, theorems) and refs/citations.yaml or hybrid references.bib. Every token label must exist; every [@citekey] must resolve.[[THMREF:…]], [[SECREF:…]], [[FIGREF:…]], [[EQREF:…]] over "Theorem 7.3", "§8", "Figure 2" in body prose.[[…]] unless renderer documents second pass.labels.yaml section/theorem numbers aligned; grep for stale numbers after renumbering.99_* uses [[CITELIST:all]], do not hand-maintain a parallel list.bashuv run python -m infrastructure.validation.cli markdown projects/<name>/manuscript/ uv run python -m infrastructure.validation.cli prerender projects/<name>/manuscript --repo-root .
docs/rules/manuscript_style.mdmanuscript/refs/README.md when present| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-11 | fail→pass | 14,222 | 1,976 | -86% | 1 | 1 | 0% | 2,086 | 1,006 | -52% | 0 | 0 | — |
case-01 | fail→fail | 4,699 | 4,773 | +2% | 1 | 1 | 0% | 256 | 1,005 | +293% | 0 | 0 | — |
case-02 | fail→fail | 4,544 | 6,115 | +35% | 1 | 1 | 0% | 360 | 1,020 | +183% | 0 | 0 | — |
case-03 | fail→fail | 5,130 | 5,768 | +12% | 1 | 1 | 0% | 303 | 950 | +214% | 0 | 0 | — |
case-04 | fail→pass | 6,358 | 3,350 | -47% | 1 | 1 | 0% | 1,185 | 1,300 | +10% | 0 | 0 | — |
case-05 | fail→pass | 9,668 | 1,597 | -83% | 1 | 1 | 0% | 1,697 | 1,026 | -40% | 0 | 0 | — |
case-06 | fail→pass | 7,572 | 4,196 | -45% | 1 | 1 | 0% | 1,436 | 1,424 | -1% | 0 | 0 | — |
case-07 | fail→pass | 6,597 | 3,975 | -40% | 1 | 1 | 0% | 1,249 | 1,486 | +19% | 0 | 0 | — |
case-08 | fail→pass | 9,211 | 4,680 | -49% | 1 | 1 | 0% | 1,403 | 1,613 | +15% | 0 | 0 | — |
case-09 | fail→pass | 4,884 | 3,267 | -33% | 1 | 1 | 0% | 802 | 1,378 | +72% | 0 | 0 | — |
case-10 | pass→pass | 9,918 | 6,405 | -35% | 1 | 1 | 0% | 1,602 | 1,868 | +17% | 0 | 0 | — |
case-12 | fail→pass | 9,913 | 3,164 | -68% | 1 | 1 | 0% | 1,673 | 1,243 | -26% | 0 | 0 | — |
case-13 | pass→pass | 8,978 | 5,781 | -36% | 1 | 1 | 0% | 1,628 | 1,778 | +9% | 0 | 0 | — |
case-14 | pass→pass | 13,067 | 9,562 | -27% | 1 | 1 | 0% | 1,876 | 2,085 | +11% | 0 | 0 | — |
case-15 | pass→pass | 5,119 | 4,052 | -21% | 1 | 1 | 0% | 837 | 1,543 | +84% | 0 | 0 | — |
case-16 | pass→pass | 9,337 | 2,734 | -71% | 1 | 1 | 0% | 1,527 | 1,276 | -16% | 0 | 0 | — |
case-17 | fail→pass | 7,030 | 2,055 | -71% | 1 | 1 | 0% | 1,162 | 1,033 | -11% | 0 | 0 | — |
case-18 | fail→fail | 16,712 | 4,811 | -71% | 1 | 1 | 0% | 2,208 | 1,574 | -29% | 0 | 0 | — |
case-19 | fail→pass | 8,958 | 2,069 | -77% | 1 | 1 | 0% | 1,341 | 1,036 | -23% | 0 | 0 | — |
case-20 | fail→fail | 15,930 | 8,347 | -48% | 1 | 1 | 0% | 2,979 | 2,289 | -23% | 0 | 0 | — |
case-21 | fail→pass | 3,927 | 6,909 | +76% | 1 | 1 | 0% | 561 | 2,026 | +261% | 0 | 0 | — |
case-22 | fail→fail | 11,889 | 8,227 | -31% | 1 | 1 | 0% | 2,307 | 2,218 | -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 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 +50 percentage points is the difference between those two pass rates over the 19 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.