Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →监管问询回复助手（IPO 版）- 协助撰写 IPO 审核问询回复。当用户需要为 IPO 项目准备交易所/证监会问询回复、梳理回复思路、组织回复材料或核查回复内容时触发此技能。适用于投行、律所、会所等中介机构的 IPO 项目场景。
.claude/skills/thomasmoreai-regulatory-inquiry-response-ipo/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-12 | ✗→✓ | ▲ Improved | 50% | 0% |
| case-04 | ✓→✓ | ＝ Same ✓ | 47% | 0% |
| case-05 | ✓→✓ | ＝ Same ✓ | 52% | 0% |
| case-06 | ✓→✓ | ＝ Same ✓ | 50% | 0% |
| case-07 | ✓→✓ | ＝ Same ✓ | 58% | 0% |
使用此技能当：
首先向用户确认以下信息：
将问询问题按以下类型分类：
为每个问题生成回复框架：
markdown## 问题 [X]：[问题标题] ### 问题原文 [交易所/证监会问询原文] ### 回复要点 1. [核心观点 1] 2. [核心观点 2] 3. [核心观点 3] ### 回复正文 #### 一、[分论点 1] [详细论述 + 数据支撑] #### 二、[分论点 2] [详细论述 + 数据支撑] #### 三、[分论点 3] [详细论述 + 数据支撑] ### 核查程序 1. [核查程序 1] 2. [核查程序 2] 3. [核查程序 3] ### 核查结论 [明确结论] ### 附件材料 - [附件 1 名称] - [附件 2 名称]
| 问题类型 | 回复策略 | |----------|----------| | 质疑持续盈利能力 | 展示增长驱动、订单储备、市场空间 | | 质疑财务真实性 | 提供核查程序、第三方证据、内控说明 | | 质疑关联交易公允性 | 对比市场价、说明必要性、披露决策程序 | | 质疑技术先进性 | 专利数量、研发投入、技术壁垒、行业认可 | | 质疑合规性 | 取得合规证明、说明整改情况、律师意见 |
为每个问题梳理所需核查材料：
markdown## 问题 [X] 核查材料清单 ### 内部文件 - [ ] 会议纪要 - [ ] 合同/协议 - [ ] 财务凭证 - [ ] 内部制度 ### 外部文件 - [ ] 客户/供应商访谈记录 - [ ] 第三方报告 - [ ] 政府证明文件 - [ ] 资质证书 ### 中介机构意见 - [ ] 保荐人核查意见 - [ ] 律师法律意见 - [ ] 会计师专项说明
提供回复质量检查清单：
references/ipo-inquiry-types.md - IPO 问询问题类型库references/response-templates.md - 回复模板库references/checklist.md - 回复质量检查清单| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 21,151 | 26,658 | +26% | 1 | 1 | 0% | 3,484 | 5,414 | +55% | 0 | 0 | — |
case-02 | fail→fail | 23,717 | 28,018 | +18% | 1 | 1 | 0% | 3,530 | 5,498 | +56% | 0 | 0 | — |
case-03 | fail→fail | 24,167 | 23,787 | -2% | 1 | 1 | 0% | 3,482 | 4,768 | +37% | 0 | 0 | — |
case-04 | pass→pass | 18,968 | 19,663 | +4% | 1 | 1 | 0% | 2,778 | 4,077 | +47% | 0 | 0 | — |
case-05 | pass→pass | 20,658 | 22,246 | +8% | 1 | 1 | 0% | 2,878 | 4,367 | +52% | 0 | 0 | — |
case-06 | pass→pass | 19,823 | 20,577 | +4% | 1 | 1 | 0% | 2,819 | 4,226 | +50% | 0 | 0 | — |
case-07 | pass→pass | 19,082 | 21,542 | +13% | 1 | 1 | 0% | 2,729 | 4,311 | +58% | 0 | 0 | — |
case-08 | pass→pass | 16,768 | 20,648 | +23% | 1 | 1 | 0% | 2,506 | 4,139 | +65% | 0 | 0 | — |
case-09 | fail→fail | 19,867 | 21,841 | +10% | 1 | 1 | 0% | 3,002 | 4,294 | +43% | 0 | 0 | — |
case-10 | pass→pass | 13,757 | 15,884 | +15% | 1 | 1 | 0% | 2,446 | 3,757 | +54% | 0 | 0 | — |
case-11 | pass→pass | 21,012 | 20,384 | -3% | 1 | 1 | 0% | 2,902 | 4,509 | +55% | 0 | 0 | — |
case-12 | fail→pass | 9,558 | 8,579 | -10% | 1 | 1 | 0% | 1,568 | 2,358 | +50% | 0 | 0 | — |
case-13 | pass→pass | 10,596 | 13,225 | +25% | 1 | 1 | 0% | 1,587 | 3,234 | +104% | 0 | 0 | — |
case-14 | pass→pass | 9,885 | 8,489 | -14% | 1 | 1 | 0% | 1,478 | 2,467 | +67% | 0 | 0 | — |
case-15 | pass→pass | 18,080 | 18,356 | +2% | 1 | 1 | 0% | 2,732 | 4,266 | +56% | 0 | 0 | — |
case-16 | pass→pass | 17,839 | 16,104 | -10% | 1 | 1 | 0% | 2,685 | 3,789 | +41% | 0 | 0 | — |
case-17 | pass→pass | 16,411 | 17,248 | +5% | 1 | 1 | 0% | 2,507 | 3,951 | +58% | 0 | 0 | — |
case-18 | pass→pass | 20,071 | 21,028 | +5% | 1 | 1 | 0% | 2,823 | 4,363 | +55% | 0 | 0 | — |
case-19 | pass→pass | 14,083 | 15,392 | +9% | 1 | 1 | 0% | 2,001 | 3,561 | +78% | 0 | 0 | — |
case-20 | pass→pass | 20,341 | 12,922 | -36% | 1 | 1 | 0% | 3,437 | 3,537 | +3% | 0 | 0 | — |
case-21 | pass→pass | 20,211 | 17,254 | -15% | 1 | 1 | 0% | 3,169 | 3,978 | +26% | 0 | 0 | — |
case-22 | pass→pass | 12,002 | 13,262 | +10% | 1 | 1 | 0% | 1,870 | 3,244 | +73% | 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 +5 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.