Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →公告解读助手，适用于资本市场、投研、财富管理、合规、行政汇报等场景。 以下情况请主动触发此技能： - 用户粘贴了任何公司公告、政府通知、监管文件、财报、问询函等文本，即使没有说"解读" - 用户问"这个算利好吗"、"这公告什么意思"、"帮我看看这个"、"是好事还是坏事"、"影响大吗"、"怎么看" - 用户提到任何公告类型：业绩预告/财报/减持/增持/回购/定增/并购/问询函/处罚/停复牌/分红/关联交易/重大合同/诉讼/风险提示 - 用户需要整理成汇报口径、研究纪要、投研摘要、领导汇报、一句话总结 - 用户描述了某件事并问"对股价有影响吗"、"对公司影响大吗"、"该怎么看"、"市场会怎么反应" - 用户提到股票代码并询问最近公告、重大事项、信息披露 不要等用户明确说"解读公告"——只要涉及公司信息披露、监管动态、资本市场事件判断、上市公司重大事项，就应主动启动此技能。即使用户只是粘贴了一段看起来像公告的文字而没有明确指令，也应该主动解读。
.claude/skills/aifinlab-announcement-interpretation-assistant/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | 26% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 91% | 0% |
| case-20 | ✗→✓ | ▲ Improved | 18% | 0% |
| case-21 | ✗→✓ | ▲ Improved | 63% | 0% |
| case-08 | ✓→✓ | ＝ Same ✓ | 49% | 0% |
你的核心职责：把"看不懂、太长、太正式"的公告，转化为"看得懂、抓重点、能判断影响"的结论。
收到用户请求后，先做两个判断：
判断1：是否有公告原文？
判断2：用户需要哪种深度？
| 用户意图 | 适用模板 | |---------|---------| | "什么意思""帮我看看""算不算利好" | 模板A：快速解读 | | "详细分析""帮我解读""影响大吗" | 模板B：标准分析 | | "汇报口径""研究纪要""投研摘要" | 模板C：汇报版 | | 未明确说明 | 默认模板A，再提供"需要详细分析可继续" |
优先级1：同花顺 iFinD
ths_report_query（需已配置登录）codes（股票代码）、reportType（如901）、beginrDate、endrDatenews_fetch_url 抓取 pdfURL 正文优先级2：用户指定来源
优先级3：联网搜索
tavily_search（或当前可用搜索工具）公司名/简称 + 公告关键词 + 日期识别公告类型后，按以下维度分析（不必每条都写，按相关性取舍）：
> 适用："什么意思""算利好吗""帮我看看"
**公告类型**：xxx
**一句话总结**：xxx
**关键信息**：
- xxx
- xxx
**初步判断**：偏[利好/利空/中性]，原因：xxx
**需关注**：xxx> 适用："详细解读""影响大吗""分析一下"
**公告概述**：xxx
**关键事实**：
- xxx
**通俗解读**：xxx（把法律措辞翻译成大白话）
**影响分析**：
- 对公司：xxx
- 对投资者：偏[利好/利空/中性]，xxx
- 主要不确定性：xxx
**后续观察点**：xxx> 适用："整理成汇报口径""研究纪要""投研摘要"
**事件概述**：xxx
**核心结论**：xxx
**关键数据与事实**：
- xxx
**影响分析**：xxx
**风险提示**：xxx
**后续跟踪**：xxx业绩预告/财报：净利润变化 + 超/低预期 + 原因（经营改善 vs 一次性因素）+ 持续性
并购重组：买谁/卖谁 + 价格与支付方式 + 协同逻辑 + 审批不确定性
减持/增持：谁在操作 + 比例节奏 + 财务性动作还是态度信号 + 市场情绪影响
回购/分红：金额/价格区间/期限 + 真实回馈还是情绪管理 + 对估值的含义
再融资/定增/可转债：融多少 + 用途 + 稀释影响 + 进攻性融资还是补流
监管问询/处罚：问题出在哪 + 财务/治理/合规影响 + 整改风险 + 经营连续性
政策通知：核心内容 + 哪些主体受影响 + 鼓励/限制/规范 + 受益/受损公司
用户问"领导汇报怎么说"：输出格式固定为 1句结论 + 3条要点 + 2条风险 + 1条建议
用户问"这公告真正重点是什么"：区分"真正影响估值/经营/预期的部分"和"只是合规表述"
公告不完整/只有摘要：直接解读，结尾标注"以下基于摘要，完整原文可能有差异"
信息不足无法判断：说明缺少哪些信息，以及拿到后能补充哪些判断，不要空转
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 15,487 | 21,088 | +36% | 1 | 1 | 0% | 1,976 | 3,452 | +75% | 0 | 0 | — |
case-02 | fail→fail | 32,105 | 29,585 | -8% | 1 | 1 | 0% | 3,911 | 5,335 | +36% | 0 | 0 | — |
case-03 | fail→fail | 27,779 | 20,274 | -27% | 1 | 1 | 0% | 3,372 | 4,016 | +19% | 0 | 0 | — |
case-04 | fail→pass | 23,583 | 16,827 | -29% | 1 | 1 | 0% | 2,753 | 3,470 | +26% | 0 | 0 | — |
case-05 | fail→pass | 13,963 | 19,337 | +38% | 1 | 1 | 0% | 1,936 | 3,706 | +91% | 0 | 0 | — |
case-06 | fail→fail | 18,187 | 29,778 | +64% | 1 | 1 | 0% | 2,706 | 5,313 | +96% | 0 | 0 | — |
case-07 | fail→fail | 38,996 | 27,939 | -28% | 1 | 1 | 0% | 3,074 | 5,645 | +84% | 0 | 0 | — |
case-08 | pass→pass | 20,248 | 21,511 | +6% | 1 | 1 | 0% | 2,973 | 4,429 | +49% | 0 | 0 | — |
case-09 | pass→pass | 44,510 | 17,482 | -61% | 1 | 1 | 0% | 3,072 | 3,831 | +25% | 0 | 0 | — |
case-10 | fail→fail | 21,282 | 27,598 | +30% | 1 | 1 | 0% | 3,029 | 4,822 | +59% | 0 | 0 | — |
case-11 | pass→pass | 20,725 | 17,795 | -14% | 1 | 1 | 0% | 2,596 | 3,866 | +49% | 0 | 0 | — |
case-12 | pass→pass | 21,163 | 19,217 | -9% | 1 | 1 | 0% | 2,926 | 4,009 | +37% | 0 | 0 | — |
case-13 | pass→pass | 17,735 | 14,492 | -18% | 1 | 1 | 0% | 2,469 | 3,215 | +30% | 0 | 0 | — |
case-14 | pass→pass | 12,251 | 27,214 | +122% | 1 | 1 | 0% | 1,644 | 4,152 | +153% | 0 | 0 | — |
case-15 | pass→pass | 9,454 | 16,354 | +73% | 1 | 1 | 0% | 1,357 | 3,381 | +149% | 0 | 0 | — |
case-16 | fail→fail | 19,779 | 15,129 | -24% | 1 | 1 | 0% | 2,564 | 3,323 | +30% | 0 | 0 | — |
case-17 | fail→fail | 37,149 | 34,651 | -7% | 1 | 1 | 0% | 6,556 | 6,136 | -6% | 0 | 0 | — |
case-18 | pass→pass | 21,489 | 12,547 | -42% | 1 | 1 | 0% | 2,988 | 3,136 | +5% | 0 | 0 | — |
case-19 | fail→fail | 20,304 | 16,855 | -17% | 1 | 1 | 0% | 2,693 | 3,808 | +41% | 0 | 0 | — |
case-20 | fail→pass | 30,420 | 22,448 | -26% | 1 | 1 | 0% | 4,082 | 4,807 | +18% | 0 | 0 | — |
case-21 | fail→pass | 15,534 | 17,543 | +13% | 1 | 1 | 0% | 2,229 | 3,623 | +63% | 0 | 0 | — |
case-22 | fail→fail | 18,014 | 15,818 | -12% | 1 | 1 | 0% | 2,233 | 3,724 | +67% | 0 | 0 | — |
case-23 | pass→pass | 24,781 | 25,456 | +3% | 1 | 1 | 0% | 3,160 | 5,021 | +59% | 0 | 0 | — |
case-24 | fail→fail | 20,735 | 8,596 | -59% | 1 | 1 | 0% | 2,622 | 2,002 | -24% | 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. 24 cases were attempted, and 23 counted toward the lift figure. The other 1 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +17 percentage points is the difference between those two pass rates over the 23 comparable cases. 3 cases got worse with the skill loaded, and they are 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.