Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use when user mentions romance, love story, or relationship-focused narrative - provides genre conventions, pacing guidelines, and emotional beats for romance writing
.claude/skills/microck-romance-novel-conventions/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 69% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 69% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 72% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 44% | 0% |
| case-11 | ✗→✓ | ▲ Improved | 46% | 0% |
| 元素 | 指导原则 | 示例 | |------|---------|------| | 初遇 | 故事的 0-10% | 意外的首次相遇，产生火花 | | 初吻 | 50-60%（慢热）或 20-30%（快热） | 需要充分的情感铺垫 | | 黑暗时刻 | 75-85% | 关系看似不可能继续 | | 大结局 | 85-100% | 表白、承诺、HEA/HFN |
| 故事阶段 | 百分比 | 关键事件 | 情感焦点 | |---------|-------|---------|---------| | 初遇 | 0-10% | 初次相遇，产生兴趣的火花 | 好奇、吸引 | | 张力建立 | 10-60% | 吸引力增长，障碍出现 | 渴望、挫折 | | 冲突升级 | 60-75% | 重大误会或真相揭露 | 怀疑、痛苦 | | 黑暗时刻 | 75-85% | 关系看似不可能 | 绝望、失落 | | 大结局 | 85-100% | 盛大表白、承诺 | 喜悦、圆满 |
慢热言情（约 50-60%）：
快热言情（约 20-30%）：
黄金法则：在没有足够铺垫的情况下，绝不要仓促。没有情感基础的身体亲密会显得空洞。
问题：角色在没有发展的情况下爱得太快
为什么不好：读者不相信这种连接；感觉勉强和不真实
解决方法：
问题："只要好好谈谈，一切都会好起来"
为什么不好：让读者沮丧；感觉像人为的戏剧
解决方法：
问题：等待被拯救或被选择；没有主动权
为什么不好：削弱角色，让言情感觉不平等
解决方法：
问题：读者已经看过一千次的可预测情节
为什么不好：无聊；读者能预测每一个转折
解决方法：
/specify 时/plan 期间/write 时/analyze 运行时写言情时，确保你有：
言情读者想要什么：
让言情读者沮丧的是什么：
记住：言情是关于两个人在一起变得比分开时更好的旅程。障碍应该迫使成长，化学反应应该感觉不可避免，结局应该感觉是赢得的。
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 23,530 | 25,073 | +7% | 1 | 1 | 0% | 3,439 | 5,825 | +69% | 0 | 0 | — |
case-02 | fail→pass | 19,311 | 17,959 | -7% | 1 | 1 | 0% | 2,631 | 4,454 | +69% | 0 | 0 | — |
case-03 | pass→pass | 20,274 | 21,017 | +4% | 1 | 1 | 0% | 2,696 | 4,759 | +77% | 0 | 0 | — |
case-04 | fail→pass | 21,019 | 21,157 | +1% | 1 | 1 | 0% | 2,845 | 4,886 | +72% | 0 | 0 | — |
case-09 | fail→pass | 20,124 | 16,007 | -20% | 1 | 1 | 0% | 2,998 | 4,327 | +44% | 0 | 0 | — |
case-05 | fail→fail | 18,566 | 20,739 | +12% | 1 | 1 | 0% | 2,542 | 4,635 | +82% | 0 | 0 | — |
case-06 | pass→pass | 22,679 | 21,631 | -5% | 1 | 1 | 0% | 3,145 | 5,060 | +61% | 0 | 0 | — |
case-07 | pass→pass | 49,024 | 19,507 | -60% | 1 | 1 | 0% | 3,338 | 4,867 | +46% | 0 | 0 | — |
case-08 | pass→pass | 57,319 | 19,842 | -65% | 1 | 1 | 0% | 2,936 | 4,874 | +66% | 0 | 0 | — |
case-10 | pass→pass | 21,757 | 22,077 | +1% | 1 | 1 | 0% | 2,984 | 4,975 | +67% | 0 | 0 | — |
case-11 | fail→pass | 23,075 | 40,885 | +77% | 1 | 1 | 0% | 3,233 | 4,736 | +46% | 0 | 0 | — |
case-12 | pass→pass | 25,172 | 22,662 | -10% | 1 | 1 | 0% | 3,275 | 5,077 | +55% | 0 | 0 | — |
case-13 | pass→pass | 19,404 | 21,723 | +12% | 1 | 1 | 0% | 2,796 | 4,891 | +75% | 0 | 0 | — |
case-14 | pass→pass | 21,071 | 18,599 | -12% | 1 | 1 | 0% | 2,779 | 4,717 | +70% | 0 | 0 | — |
case-15 | pass→pass | 21,204 | 18,874 | -11% | 1 | 1 | 0% | 2,893 | 4,524 | +56% | 0 | 0 | — |
case-16 | pass→pass | 13,479 | 5,747 | -57% | 1 | 1 | 0% | 1,944 | 2,617 | +35% | 0 | 0 | — |
case-17 | pass→fail | 20,497 | 18,809 | -8% | 1 | 1 | 0% | 2,944 | 4,718 | +60% | 0 | 0 | — |
case-18 | pass→pass | 19,960 | 20,237 | +1% | 1 | 1 | 0% | 2,617 | 4,756 | +82% | 0 | 0 | — |
case-19 | fail→pass | 14,155 | 8,792 | -38% | 1 | 1 | 0% | 2,344 | 3,257 | +39% | 0 | 0 | — |
case-20 | pass→pass | 26,678 | 25,497 | -4% | 1 | 1 | 0% | 4,052 | 5,960 | +47% | 0 | 0 | — |
case-21 | pass→pass | 17,852 | 18,951 | +6% | 1 | 1 | 0% | 3,715 | 5,685 | +53% | 0 | 0 | — |
case-22 | pass→pass | 30,230 | 36,026 | +19% | 1 | 1 | 0% | 4,288 | 6,772 | +58% | 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 +23 percentage points is the difference between those two pass rates over the 22 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
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.