Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use when deciding which cpa-* sub-skill to invoke next, or when sequencing manuscript work from topic selection through rebuttal for 《中国行政管理》 (Chinese Public Administration). Routes — does not replace — the specialized skills.
.claude/skills/brycewang-stanford-cpa-workflow/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 19% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 37% | 0% |
| case-22 | ✗→✓ | ▲ Improved | 9% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -3% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -9% | 0% |
这是路由器，它不替代任何专用 Skill。它告诉你当前阶段应该用哪一个 cpa- skill。
默认假设：除非用户明确说明目标期刊不是《中国行政管理》，否则按中国行政管理学会主办、CSSCI 来源期刊《中国行政管理》的编委口味处理——公共行政 / 公共管理学科，方法上定量与定性并重，须有公共管理理论框架。
| 当前症状 | 下一个 Skill | |----------------------------------------------|---------------------------| | 选题宏大空泛，缺公共管理理论定位 / 落不成可研究的问题 | cpa-topic-selection | | 综述缺公共管理理论文献 / 缺中国治理文献 / 罗列式 | cpa-literature-review | | 研究设计与问题不匹配 / 方法粗糙（仅描述性 + 主观判断） | cpa-identification | | 有结论但缺治理 / 政策过程机制 | cpa-mechanism | | 没切跨地区 / 层级 / 治理情境差异 | cpa-heterogeneity | | 表格列数过多 / 编码表缺失 / 机制图不规范 | cpa-tables-figures | | 文末是"强化 / 完善 / 健全 / 推进"四件套或空泛对策 | cpa-policy-implication | | 摘要写成套话 / 没说清问题方法发现 / 中英不对齐 | cpa-abstract | | 通篇对策套话、无理论支撑 / 命中黑名单短语 | cpa-style | | 准备投稿，需要 checklist | cpa-submission | | 收到外审意见，需要写回复信 | cpa-rebuttal |
cpa-topic-selection — 先把"理论框架 + 可研究的中国治理问题"定下来cpa-literature-review — 公共管理理论与中国治理文献并重，对话式综述cpa-identification — 研究设计：定量 / 定性 / 规范，方法与问题匹配cpa-mechanism — 治理 / 政策过程机制cpa-heterogeneity — 跨情境差异 / 条件性cpa-tables-figures — 主表 / 编码表 / 机制图最终化cpa-policy-implication — 政策含义（可操作但有据）cpa-abstract — 摘要四段法 + 黑名单短语清除cpa-style — 全文语言风格 polishcpa-submission — 投稿前 preflightcpa-rebuttal — 外审后> cpa-abstract 与 cpa-style 是后段 polish 阶段触发，不要在研究设计未立住时去做。
不确定卡在哪一环时，逐题作答；遇到第一个"否"就停，右侧即下一个 skill：
text【《中国行政管理》路由诊断卡】 Q1 理论框架能用一句话说出（框架名=____）？ 否 → cpa-topic-selection Q2 综述在与公共管理理论"对话"而非罗列？ 否 → cpa-literature-review Q3 方法与问题匹配（定量/定性/规范，选=____）？ 否 → cpa-identification Q4 治理或政策过程的机制黑箱已打开？ 否 → cpa-mechanism Q5 跨地区/层级/治理情境的条件性已分析？ 否 → cpa-heterogeneity Q6 主表、资料编码表、机制图均已规范？ 否 → cpa-tables-figures Q7 政策建议有分析支撑且可操作？ 否 → cpa-policy-implication Q8 摘要含"问题—方法—发现"三要素？ 否 → cpa-abstract Q9 全文黑名单短语命中 0 处？ 否 → cpa-style Q10 投稿 checklist 全部通过？ 否 → cpa-submission （已收到匿名评审意见 → 跳过 Q1–Q10，直接 cpa-rebuttal）
cpa-topic-selectioncpa-literature-reviewcpa-identificationcpa-mechanismcpa-heterogeneitycpa-tables-figurescpa-policy-implicationcpa-abstractcpa-stylecpa-submissioncpa-rebuttal如果稿子本质是经济学（强调因果识别、经济理论、政策意义层启示），用 Economic-Research-Journal-Skills 更合适。两者核心差异：
cpa-topic-selection 直接进研究设计——审稿人首先看理论定位与问题是否可研究cpa-policy-implication 在没有分析支撑时写"对策四件套"cpa-rebuttal 在你修订正文之前生成回复信| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 28,480 | 17,667 | -38% | 1 | 1 | 0% | 3,018 | 3,590 | +19% | 0 | 0 | — |
case-02 | pass→pass | 22,649 | 17,274 | -24% | 1 | 1 | 0% | 2,545 | 3,508 | +38% | 0 | 0 | — |
case-03 | fail→pass | 24,748 | 17,311 | -30% | 1 | 1 | 0% | 2,571 | 3,524 | +37% | 0 | 0 | — |
case-22 | fail→pass | 22,195 | 10,614 | -52% | 1 | 1 | 0% | 2,099 | 2,291 | +9% | 0 | 0 | — |
case-04 | fail→pass | 22,280 | 11,704 | -47% | 1 | 1 | 0% | 2,518 | 2,454 | -3% | 0 | 0 | — |
case-05 | fail→pass | 19,341 | 13,287 | -31% | 1 | 1 | 0% | 2,881 | 2,617 | -9% | 0 | 0 | — |
case-06 | pass→pass | 13,016 | 5,393 | -59% | 1 | 1 | 0% | 1,595 | 2,260 | +42% | 0 | 0 | — |
case-07 | fail→pass | 21,541 | 14,430 | -33% | 1 | 1 | 0% | 2,421 | 2,832 | +17% | 0 | 0 | — |
case-08 | fail→pass | 25,695 | 11,071 | -57% | 1 | 1 | 0% | 2,621 | 2,574 | -2% | 0 | 0 | — |
case-09 | fail→pass | 20,899 | 12,263 | -41% | 1 | 1 | 0% | 2,049 | 2,463 | +20% | 0 | 0 | — |
case-10 | fail→pass | 18,564 | 10,798 | -42% | 1 | 1 | 0% | 1,839 | 2,274 | +24% | 0 | 0 | — |
case-11 | fail→pass | 27,312 | 12,748 | -53% | 1 | 1 | 0% | 2,471 | 2,719 | +10% | 0 | 0 | — |
case-12 | pass→pass | 25,489 | 16,898 | -34% | 1 | 1 | 0% | 2,450 | 3,279 | +34% | 0 | 0 | — |
case-13 | pass→pass | 26,620 | 19,241 | -28% | 1 | 1 | 0% | 2,621 | 3,825 | +46% | 0 | 0 | — |
case-14 | fail→pass | 38,160 | 25,667 | -33% | 1 | 1 | 0% | 3,194 | 4,148 | +30% | 0 | 0 | — |
case-15 | pass→pass | 24,819 | 23,192 | -7% | 1 | 1 | 0% | 2,399 | 3,760 | +57% | 0 | 0 | — |
case-16 | pass→pass | 23,231 | 19,625 | -16% | 1 | 1 | 0% | 2,108 | 3,364 | +60% | 0 | 0 | — |
case-21 | fail→pass | 19,574 | 11,395 | -42% | 1 | 1 | 0% | 1,982 | 2,383 | +20% | 0 | 0 | — |
case-17 | pass→pass | 21,941 | 13,395 | -39% | 1 | 1 | 0% | 2,106 | 3,562 | +69% | 0 | 0 | — |
case-18 | pass→pass | 17,889 | 24,838 | +39% | 1 | 1 | 0% | 2,318 | 4,262 | +84% | 0 | 0 | — |
case-19 | pass→pass | 23,202 | 22,909 | -1% | 1 | 1 | 0% | 2,443 | 4,002 | +64% | 0 | 0 | — |
case-20 | pass→fail | 29,251 | 38,859 | +33% | 1 | 1 | 0% | 3,054 | 5,325 | +74% | 0 | 0 | — |
case-23 | fail→pass | 27,902 | 8,792 | -68% | 1 | 1 | 0% | 2,783 | 2,185 | -21% | 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. 23 cases were attempted. The headline lift of +52 percentage points is the difference between those two pass rates over the 23 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.