Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →当用户提到悬疑、侦探、犯罪或悬念叙事时使用 - 提供类型规范、线索布置和推理小说的公平游戏原则
.claude/skills/microck-mystery-novel-conventions/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | 85% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 48% | 0% |
| case-12 | ✗→✓ | ▲ Improved | 68% | 0% |
| case-18 | ✗→✓ | ▲ Improved | 118% | 0% |
| case-01 | ✓→✓ | ＝ Same ✓ | 103% | 0% |
| 元素 | 指导原则 | 位置 | | ------------ | -------------- | ---------------- | | 触发事件 | 案件/谜团发生 | 前 10% | | 误导线索 | 错误的引导 | 贯穿全文，3-5 个 | | 真实线索 | 公平游戏的证据 | 75% 之前 | | 真相揭露 | 真相揭示 | 85-95% | | 收尾 | 结束所有线索 | 最后 5% |
黄金法则：读者必须在侦探之前获得所有解决谜团所需的线索。
建立常态世界：
案件发生：
初步调查：
收集线索：
误导线索：
复杂化升级：
中点转折（约 50%）：
黑暗之夜（75-85%）：
真相揭露（85-95%）：
收尾（95-100%）：
物理证据：
证言证据：
行为证据：
circumstantial证据：
早期线索（0-25%）：
中期线索（25-75%）：
后期线索（75-85%）：
特征：
示例：
❌ 太明显：读者立即看穿 ❌ 太勉强：感觉强迫和人为 ❌ 从不解释：悬而未决没有解决 ❌ 太多：读者失去追踪并感到沮丧
至少 3 个嫌疑人：
罪犯：
误导嫌疑人：
同情嫌疑人：
每个嫌疑人都应该有秘密：
问题：罪犯或方法依赖读者没有的信息
解决：在 75% 标记之前埋下所有必要线索；读者应该能够解决
问题：主角错过明显线索或行为不合逻辑
解决：让侦探有能力但人性化；他们可以犯错，但不是愚蠢
问题：情节通过方便的运气而非调查推进
解决：巧合可以使情况复杂化，永远不能解决；侦探必须努力寻找答案
问题：调查变成重复的访谈接访谈
解决：变化调查方法；添加动作、危险、个人风险
问题：复杂的解决方案在最后一章的对话中倾倒
解决：分散揭示；让读者拼凑；保持解释清晰但不冗长
/specify 执行时/plan 期间/write 时/analyze 期间层次 1 - 明显（25% 的线索）：
层次 2 - 微妙（50% 的线索）：
层次 3 - 藏在显眼处（25% 的线索）：
层次 1："窗户从里面解锁" 层次 2：角色在闲聊中提到自己是左撇子 层次 3：房间描述包括烟灰缸中的特定品牌香烟
悬疑读者想要什么：
让悬疑读者沮丧的是什么：
记住：一个伟大的谜团让读者因解决它而感到聪明，或因没有看到它而印象深刻 - 但总是满意线索一直都在那里。
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | pass→pass | 15,569 | 13,763 | -12% | 1 | 1 | 0% | 2,239 | 4,552 | +103% | 0 | 0 | — |
case-02 | fail→fail | 14,604 | 12,615 | -14% | 1 | 1 | 0% | 2,091 | 4,398 | +110% | 0 | 0 | — |
case-03 | pass→pass | 12,216 | 11,093 | -9% | 1 | 1 | 0% | 1,808 | 4,147 | +129% | 0 | 0 | — |
case-04 | fail→pass | 12,886 | 7,685 | -40% | 1 | 1 | 0% | 1,990 | 3,681 | +85% | 0 | 0 | — |
case-05 | pass→pass | 14,399 | 14,127 | -2% | 1 | 1 | 0% | 2,252 | 4,453 | +98% | 0 | 0 | — |
case-06 | pass→pass | 13,085 | 4,221 | -68% | 1 | 1 | 0% | 2,075 | 3,078 | +48% | 0 | 0 | — |
case-07 | fail→pass | 13,403 | 3,903 | -71% | 1 | 1 | 0% | 2,130 | 3,143 | +48% | 0 | 0 | — |
case-08 | pass→pass | 12,947 | 13,378 | +3% | 1 | 1 | 0% | 1,962 | 4,471 | +128% | 0 | 0 | — |
case-09 | pass→pass | 14,800 | 12,833 | -13% | 1 | 1 | 0% | 2,088 | 4,355 | +109% | 0 | 0 | — |
case-10 | pass→pass | 11,309 | 5,251 | -54% | 1 | 1 | 0% | 1,674 | 3,178 | +90% | 0 | 0 | — |
case-11 | pass→pass | 12,336 | 4,203 | -66% | 1 | 1 | 0% | 1,800 | 3,073 | +71% | 0 | 0 | — |
case-12 | fail→pass | 10,452 | 3,398 | -67% | 1 | 1 | 0% | 1,763 | 2,963 | +68% | 0 | 0 | — |
case-13 | pass→pass | 16,498 | 7,804 | -53% | 1 | 1 | 0% | 2,384 | 3,787 | +59% | 0 | 0 | — |
case-14 | pass→pass | 15,909 | 17,304 | +9% | 1 | 1 | 0% | 2,377 | 5,117 | +115% | 0 | 0 | — |
case-15 | pass→pass | 15,086 | 15,681 | +4% | 1 | 1 | 0% | 2,320 | 4,864 | +110% | 0 | 0 | — |
case-16 | pass→pass | 10,244 | 3,374 | -67% | 1 | 1 | 0% | 1,713 | 2,975 | +74% | 0 | 0 | — |
case-17 | pass→pass | 15,367 | 9,651 | -37% | 1 | 1 | 0% | 2,310 | 4,062 | +76% | 0 | 0 | — |
case-18 | fail→pass | 14,781 | 14,689 | -1% | 1 | 1 | 0% | 2,218 | 4,825 | +118% | 0 | 0 | — |
case-19 | pass→pass | 16,426 | 15,374 | -6% | 1 | 1 | 0% | 2,395 | 4,787 | +100% | 0 | 0 | — |
case-20 | pass→pass | 25,995 | 27,407 | +5% | 1 | 1 | 0% | 3,556 | 6,169 | +73% | 0 | 0 | — |
case-21 | pass→pass | 16,502 | 13,726 | -17% | 1 | 1 | 0% | 2,444 | 4,627 | +89% | 0 | 0 | — |
case-22 | pass→pass | 18,454 | 15,740 | -15% | 1 | 1 | 0% | 2,687 | 4,982 | +85% | 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 +18 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.