Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →对中文毕业论文做最终润色与去 AI 味：重点处理章首导言、章末小结、贡献描述、总结与展望中的模板腔、宣传腔与 AI 口癖。 **Trigger**: 论文润色, thesis style polish, 去 AI 味, 中文论文风格统一, 章首导言润色, 总结展望润色. **Use when**: 结构、证据、数据和编译已经基本稳定，需要做最后一轮中文学位论文风格收口。 **Skip if**: 正文结构还在大改，或关键引用 / 数字 / 图表还没稳定。 **Network**: none. **Guardrail**: 不用润色掩盖结构问题；不引入新事实；先去模板腔，再做措辞提升。
.claude/skills/willoscar-thesis-style-polisher/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | -21% | 0% |
| case-02 | ✗→✓ | ▲ Improved | -67% | 0% |
| case-03 | ✗→✓ | ▲ Improved | -50% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -50% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -72% | 0% |
中文写作要求.mdGPT口癖与高频用词调研.mdAlways read:
references/overview.mdreferences/style_principles.mdreferences/examples_good.mdreferences/examples_bad.mdMachine-readable contract:
assets/style_polish_contract.json如果一段话的问题本质上是“没想清楚”，不要在这里强行润色。
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 13,767 | 9,166 | -33% | 1 | 1 | 0% | 1,872 | 1,478 | -21% | 0 | 0 | — |
case-02 | fail→pass | 11,542 | 2,251 | -80% | 1 | 1 | 0% | 1,626 | 542 | -67% | 0 | 0 | — |
case-03 | fail→pass | 11,892 | 4,506 | -62% | 1 | 1 | 0% | 1,603 | 802 | -50% | 0 | 0 | — |
case-04 | pass→pass | 14,344 | 4,358 | -70% | 1 | 1 | 0% | 1,792 | 743 | -59% | 0 | 0 | — |
case-05 | fail→pass | 6,204 | 1,844 | -70% | 1 | 1 | 0% | 817 | 408 | -50% | 0 | 0 | — |
case-06 | fail→pass | 12,416 | 2,088 | -83% | 1 | 1 | 0% | 1,735 | 480 | -72% | 0 | 0 | — |
case-07 | fail→pass | 9,167 | 1,964 | -79% | 1 | 1 | 0% | 1,396 | 465 | -67% | 0 | 0 | — |
case-08 | fail→pass | 7,607 | 1,789 | -76% | 1 | 1 | 0% | 1,093 | 418 | -62% | 0 | 0 | — |
case-09 | fail→pass | 8,674 | 1,906 | -78% | 1 | 1 | 0% | 1,220 | 430 | -65% | 0 | 0 | — |
case-10 | fail→pass | 9,995 | 1,739 | -83% | 1 | 1 | 0% | 1,438 | 412 | -71% | 0 | 0 | — |
case-11 | fail→pass | 14,163 | 2,153 | -85% | 1 | 1 | 0% | 1,983 | 469 | -76% | 0 | 0 | — |
case-12 | pass→pass | 9,968 | 2,615 | -74% | 1 | 1 | 0% | 1,348 | 602 | -55% | 0 | 0 | — |
case-13 | pass→pass | 10,267 | 4,237 | -59% | 1 | 1 | 0% | 1,455 | 760 | -48% | 0 | 0 | — |
case-14 | pass→pass | 9,368 | 3,841 | -59% | 1 | 1 | 0% | 1,211 | 747 | -38% | 0 | 0 | — |
case-15 | pass→pass | 11,925 | 3,423 | -71% | 1 | 1 | 0% | 1,591 | 652 | -59% | 0 | 0 | — |
case-16 | fail→pass | 12,939 | 6,964 | -46% | 1 | 1 | 0% | 1,665 | 1,084 | -35% | 0 | 0 | — |
case-17 | fail→pass | 7,086 | 2,668 | -62% | 1 | 1 | 0% | 980 | 557 | -43% | 0 | 0 | — |
case-18 | fail→pass | 9,642 | 3,126 | -68% | 1 | 1 | 0% | 1,379 | 686 | -50% | 0 | 0 | — |
case-19 | fail→pass | 7,073 | 3,104 | -56% | 1 | 1 | 0% | 946 | 684 | -28% | 0 | 0 | — |
case-20 | fail→pass | 13,286 | 10,777 | -19% | 1 | 1 | 0% | 1,873 | 1,522 | -19% | 0 | 0 | — |
case-21 | pass→pass | 6,719 | 5,925 | -12% | 1 | 1 | 0% | 1,038 | 1,124 | +8% | 0 | 0 | — |
case-22 | pass→pass | 6,967 | 6,601 | -5% | 1 | 1 | 0% | 1,037 | 996 | -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. The headline lift of +68 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.