Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Compile the Beamer tutorial deck and write a build report. **Trigger**: beamer compile, slides compile, tutorial slides pdf, 编译幻灯片, beamer pdf. **Use when**: `source-tutorial` 的 C4，已有 `latex/slides/main.tex`，需要输出 `latex/slides/main.pdf` 和编译报告。 **Skip if**: slides scaffold 还没完成。 **Network**: none. **Guardrail**: 编译失败也要落盘可读报告，不能只返回报错。
.claude/skills/willoscar-beamer-compile-qa/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-09 | ✗→✓ | ▲ Improved | -71% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -60% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -28% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -59% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -7% | 0% |
source-tutorial 的 C4，已有 latex/slides/main.tex，需要输出 latex/slides/main.pdf 和编译报告。Goal: compile the Beamer deck and record success or failure in a stable report.
The compiler should treat latex/slides/main.tex as the entrypoint and emit both latex/slides/main.pdf and output/SLIDES_BUILD_REPORT.md.
latex/slides/main.texlatex/slides/main.pdfoutput/SLIDES_BUILD_REPORT.mduv run python .codex/skills/beamer-compile-qa/scripts/run.py --workspace <workspace>--workspace <dir> (required)--unit-id <U###>--inputs <semicolon-separated>--outputs <semicolon-separated>--checkpoint <C#>uv run python .codex/skills/beamer-compile-qa/scripts/run.py --workspace <workspace>| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-03 | fail→fail | 5,721 | 5,059 | -12% | 1 | 1 | 0% | 216 | 426 | +97% | 0 | 0 | — |
case-09 | fail→pass | 12,702 | 2,354 | -81% | 1 | 1 | 0% | 2,252 | 642 | -71% | 0 | 0 | — |
case-01 | fail→fail | 11,931 | 5,426 | -55% | 1 | 1 | 0% | 1,980 | 537 | -73% | 0 | 0 | — |
case-02 | fail→fail | 4,793 | 5,304 | +11% | 1 | 1 | 0% | 201 | 466 | +132% | 0 | 0 | — |
case-04 | fail→pass | 8,626 | 2,078 | -76% | 1 | 1 | 0% | 1,399 | 558 | -60% | 0 | 0 | — |
case-05 | fail→pass | 5,614 | 2,446 | -56% | 1 | 1 | 0% | 918 | 665 | -28% | 0 | 0 | — |
case-06 | fail→fail | 4,412 | 4,780 | +8% | 1 | 1 | 0% | 198 | 441 | +123% | 0 | 0 | — |
case-07 | fail→pass | 11,297 | 3,666 | -68% | 1 | 1 | 0% | 1,911 | 785 | -59% | 0 | 0 | — |
case-08 | fail→pass | 19,206 | 5,239 | -73% | 1 | 1 | 0% | 1,194 | 1,107 | -7% | 0 | 0 | — |
case-10 | pass→pass | 6,019 | 2,260 | -62% | 1 | 1 | 0% | 968 | 588 | -39% | 0 | 0 | — |
case-11 | fail→pass | 4,562 | 1,723 | -62% | 1 | 1 | 0% | 637 | 460 | -28% | 0 | 0 | — |
case-12 | fail→pass | 7,363 | 1,507 | -80% | 1 | 1 | 0% | 1,050 | 434 | -59% | 0 | 0 | — |
case-13 | fail→pass | 8,005 | 1,627 | -80% | 1 | 1 | 0% | 1,184 | 445 | -62% | 0 | 0 | — |
case-14 | fail→pass | 16,160 | 1,381 | -91% | 1 | 1 | 0% | 1,239 | 411 | -67% | 0 | 0 | — |
case-15 | fail→pass | 18,322 | 1,639 | -91% | 1 | 1 | 0% | 1,388 | 426 | -69% | 0 | 0 | — |
case-16 | pass→pass | 7,825 | 1,619 | -79% | 1 | 1 | 0% | 1,173 | 453 | -61% | 0 | 0 | — |
case-17 | pass→pass | 14,848 | 1,277 | -91% | 1 | 1 | 0% | 2,345 | 372 | -84% | 0 | 0 | — |
case-18 | fail→pass | 15,234 | 1,525 | -90% | 1 | 1 | 0% | 2,434 | 400 | -84% | 0 | 0 | — |
case-19 | fail→pass | 7,667 | 7,841 | +2% | 1 | 1 | 0% | 1,193 | 1,042 | -13% | 0 | 0 | — |
case-20 | fail→fail | 2,928 | 4,252 | +45% | 1 | 1 | 0% | 390 | 426 | +9% | 0 | 0 | — |
case-21 | fail→fail | 7,117 | 4,458 | -37% | 1 | 1 | 0% | 1,231 | 418 | -66% | 0 | 0 | — |
case-22 | pass→pass | 8,320 | 6,294 | -24% | 1 | 1 | 0% | 1,443 | 1,278 | -11% | 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 15 counted toward the lift figure. The other 7 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 +55 percentage points is the difference between those two pass rates over the 15 comparable cases.
The publisher has shipped newer versions since this run, so these numbers describe v1, not the version currently listed.
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.