Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →A股聪明钱指标/主力行为量化识别。当用户说"聪明钱"、"smart money"、"主力行为"、"大单分析"、"机构行为"、"量价背离"、"主力在干嘛"、"有没有主力进场"时触发。基于 cn-stock-data 获取资金流向和K线数据，量化识别机构/主力资金行为模式。支持研报风格（formal）和快速分析风格（brief）。
.claude/skills/aifinlab-a-share-smart-money/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-17 | ✗→✓ | ▲ Improved | 8% | 0% |
| case-19 | ✗→✓ | ▲ Improved | 3% | 0% |
| case-20 | ✗→✓ | ▲ Improved | 5% | 0% |
| case-07 | ✓→✓ | ＝ Same ✓ | 10% | 0% |
| case-08 | ✓→✓ | ＝ Same ✓ | 16% | 0% |
通过 cn-stock-data 统一层获取：
大单净流入 / 总成交额 × 100%CLV = ((close-low)-(high-close))/(high-low), AD += CLV × volume| 模式 | 价格特征 | 量能特征 | 资金流特征 | |------|---------|---------|-----------| | 吸筹 accumulation | 低位横盘/缓跌 | 缩量后间歇放量 | 大单持续净流入 | | 出货 distribution | 高位震荡/滞涨 | 高位放量 | 大单持续净流出 | | 洗盘 washout | 急跌后快速收回 | 下跌缩量反弹放量 | 大单流出但占比小 | | 拉升 markup | 连续上涨/突破 | 持续放量 | 大单强劲流入 |
scripts/smart_money_detector.py 计算指标bash# 准备数据后调用 python scripts/smart_money_detector.py --flow flow.json --kline kline.json # 可选: --days 20 (分析窗口，默认20)
python# 调用 skill result = run_skill({ "param1": "value1", "param2": "value2" })
bashpython scripts/run_skill.py --input data.json
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 92,059 | 8,710 | -91% | 1 | 1 | 0% | 4,628 | 1,410 | -70% | 0 | 0 | — |
case-02 | fail→fail | 12,851 | 10,669 | -17% | 1 | 1 | 0% | 1,718 | 2,695 | +57% | 0 | 0 | — |
case-03 | fail→fail | 41,432 | 11,539 | -72% | 1 | 1 | 0% | 3,191 | 1,468 | -54% | 0 | 0 | — |
case-04 | fail→fail | 25,122 | 30,158 | +20% | 1 | 1 | 0% | 4,286 | 6,067 | +42% | 0 | 0 | — |
case-05 | fail→fail | 21,150 | 37,062 | +75% | 1 | 1 | 0% | 3,938 | 5,137 | +30% | 0 | 0 | — |
case-06 | fail→fail | 26,025 | 32,004 | +23% | 1 | 1 | 0% | 3,243 | 5,205 | +60% | 0 | 0 | — |
case-07 | pass→pass | 14,669 | 9,562 | -35% | 1 | 1 | 0% | 2,316 | 2,545 | +10% | 0 | 0 | — |
case-08 | pass→pass | 23,630 | 17,260 | -27% | 1 | 1 | 0% | 3,103 | 3,599 | +16% | 0 | 0 | — |
case-09 | pass→pass | 18,077 | 22,764 | +26% | 1 | 1 | 0% | 2,697 | 3,653 | +35% | 0 | 0 | — |
case-10 | pass→pass | 30,211 | 10,532 | -65% | 1 | 1 | 0% | 3,098 | 2,887 | -7% | 0 | 0 | — |
case-11 | pass→pass | 18,085 | 15,657 | -13% | 1 | 1 | 0% | 2,510 | 3,203 | +28% | 0 | 0 | — |
case-12 | pass→pass | 18,555 | 14,152 | -24% | 1 | 1 | 0% | 2,546 | 2,743 | +8% | 0 | 0 | — |
case-13 | pass→pass | 20,004 | 16,439 | -18% | 1 | 1 | 0% | 3,003 | 3,345 | +11% | 0 | 0 | — |
case-14 | pass→pass | 24,692 | 20,166 | -18% | 1 | 1 | 0% | 2,914 | 3,835 | +32% | 0 | 0 | — |
case-15 | pass→pass | 19,444 | 17,934 | -8% | 1 | 1 | 0% | 3,035 | 3,628 | +20% | 0 | 0 | — |
case-16 | pass→pass | 24,228 | 21,482 | -11% | 1 | 1 | 0% | 2,977 | 3,800 | +28% | 0 | 0 | — |
case-17 | fail→pass | 8,723 | 2,577 | -70% | 1 | 1 | 0% | 1,269 | 1,371 | +8% | 0 | 0 | — |
case-18 | pass→pass | 12,332 | 2,862 | -77% | 1 | 1 | 0% | 1,560 | 1,295 | -17% | 0 | 0 | — |
case-19 | fail→pass | 20,019 | 9,837 | -51% | 1 | 1 | 0% | 2,632 | 2,698 | +3% | 0 | 0 | — |
case-20 | fail→pass | 7,686 | 3,152 | -59% | 1 | 1 | 0% | 1,256 | 1,323 | +5% | 0 | 0 | — |
case-21 | fail→fail | 22,352 | 13,381 | -40% | 1 | 1 | 0% | 2,816 | 2,836 | +1% | 0 | 0 | — |
case-22 | pass→pass | 18,966 | 14,095 | -26% | 1 | 1 | 0% | 3,573 | 2,593 | -27% | 0 | 0 | — |
case-23 | pass→pass | 29,306 | 18,053 | -38% | 1 | 1 | 0% | 3,276 | 3,765 | +15% | 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, and 21 counted toward the lift figure. The other 2 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 +13 percentage points is the difference between those two pass rates over the 21 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.