Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →把已经在 Markdown 中间层收敛的中文毕业论文内容回写到 `chapters/*.tex` 与相关交付层文件，保持结构、图表、公式、交叉引用与章节承接一致。 **Trigger**: tex 回写, thesis tex writeback, md 回 tex, 章节回写, 论文交付层同步. **Use when**: 章节主线、术语、符号和图表计划已经在 Markdown 层基本稳定，需要进入 TeX 交付层。 **Skip if**: 结构还没稳定，或者仍在章级重构阶段。 **Network**: none. **Guardrail**: 不在这里重新发明结构；优先把结构问题回到 Markdown 层处理；TeX 层负责交付而非重新思考。
.claude/skills/willoscar-thesis-tex-writeback/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-09 | ✗→✓ | ▲ Improved | -51% | 0% |
| case-18 | ✗→✓ | ▲ Improved | -7% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -40% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -29% | 0% |
| case-11 | ✗→✓ | ▲ Improved | 11% | 0% |
codex_md/00_thesis_outline.mdcodex_md/terminology_glossary.mdcodex_md/symbol_metric_table.mdcodex_md/figure_plan.mdchapters/*.texchapters/*.texAlways read:
references/overview.mdreferences/writeback_rules.mdreferences/consistency_checks.mdMachine-readable contract:
assets/tex_writeback_contract.jsonchapters/*.tex如果问题本质上属于：
就不应该留在 tex 层解决。
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-09 | fail→pass | 8,371 | 1,922 | -77% | 1 | 1 | 0% | 1,148 | 564 | -51% | 0 | 0 | — |
case-18 | fail→pass | 5,451 | 2,376 | -56% | 1 | 1 | 0% | 739 | 684 | -7% | 0 | 0 | — |
case-01 | fail→fail | 6,357 | 5,203 | -18% | 1 | 1 | 0% | 941 | 508 | -46% | 0 | 0 | — |
case-02 | fail→fail | 4,672 | 5,266 | +13% | 1 | 1 | 0% | 685 | 550 | -20% | 0 | 0 | — |
case-03 | fail→fail | 7,798 | 4,454 | -43% | 1 | 1 | 0% | 1,007 | 503 | -50% | 0 | 0 | — |
case-04 | fail→pass | 7,869 | 2,555 | -68% | 1 | 1 | 0% | 1,133 | 684 | -40% | 0 | 0 | — |
case-05 | pass→pass | 8,804 | 3,673 | -58% | 1 | 1 | 0% | 1,378 | 898 | -35% | 0 | 0 | — |
case-06 | fail→pass | 13,515 | 6,426 | -52% | 1 | 1 | 0% | 1,821 | 1,296 | -29% | 0 | 0 | — |
case-07 | pass→pass | 9,550 | 4,550 | -52% | 1 | 1 | 0% | 1,427 | 1,021 | -28% | 0 | 0 | — |
case-08 | pass→pass | 10,318 | 5,284 | -49% | 1 | 1 | 0% | 1,469 | 1,132 | -23% | 0 | 0 | — |
case-10 | fail→fail | 12,045 | 12,917 | +7% | 1 | 1 | 0% | 2,004 | 2,360 | +18% | 0 | 0 | — |
case-11 | fail→pass | 11,994 | 11,324 | -6% | 1 | 1 | 0% | 1,750 | 1,943 | +11% | 0 | 0 | — |
case-12 | fail→fail | 8,242 | 5,070 | -38% | 1 | 1 | 0% | 1,128 | 1,003 | -11% | 0 | 0 | — |
case-13 | pass→pass | 13,725 | 8,059 | -41% | 1 | 1 | 0% | 2,239 | 1,433 | -36% | 0 | 0 | — |
case-14 | pass→pass | 8,342 | 7,017 | -16% | 1 | 1 | 0% | 1,211 | 1,350 | +11% | 0 | 0 | — |
case-15 | pass→pass | 10,517 | 5,712 | -46% | 1 | 1 | 0% | 1,464 | 1,105 | -25% | 0 | 0 | — |
case-16 | fail→pass | 10,397 | 2,850 | -73% | 1 | 1 | 0% | 1,572 | 696 | -56% | 0 | 0 | — |
case-17 | fail→fail | 7,496 | 4,722 | -37% | 1 | 1 | 0% | 1,046 | 946 | -10% | 0 | 0 | — |
case-19 | fail→pass | 12,384 | 9,730 | -21% | 1 | 1 | 0% | 1,803 | 1,839 | +2% | 0 | 0 | — |
case-20 | pass→pass | 13,143 | 14,702 | +12% | 1 | 1 | 0% | 2,118 | 2,484 | +17% | 0 | 0 | — |
case-21 | pass→pass | 22,049 | 20,948 | -5% | 1 | 1 | 0% | 4,170 | 4,243 | +2% | 0 | 0 | — |
case-22 | pass→pass | 5,401 | 3,063 | -43% | 1 | 1 | 0% | 884 | 760 | -14% | 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 +32 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.