Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →A股交易信号回测/策略验证。当用户说"信号回测"、"signal backtest"、"这个信号准不准"、"信号验证"、"交易信号回测"时触发。量化回测特定交易信号的历史表现。支持formal和brief风格。
.claude/skills/aifinlab-a-share-signal-backtest/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | -52% | 0% |
| case-09 | ✗→✓ | ▲ Improved | -27% | 0% |
| case-10 | ✗→✓ | ▲ Improved | -50% | 0% |
| case-11 | ✗→✓ | ▲ Improved | -55% | 0% |
| case-14 | ✗→✓ | ▲ Improved | -24% | 0% |
bashSCRIPTS="$SKILLS_ROOT/cn-stock-data/scripts" python "$SCRIPTS/cn_stock_data.py" kline --code [CODE] --freq daily --start [日期] python "$SCRIPTS/cn_stock_data.py" quote --code [CODE] python "$SCRIPTS/cn_stock_data.py" finance --code [CODE]
明确入场/出场条件
标记每个交易日的信号（买入/卖出/持有）
| 维度 | formal | brief | |------|--------|-------| | 绩效 | 完整回测报告 | 胜率+夏普 | | 交易明细 | 每笔交易记录 | 统计摘要 | | 稳健性 | 分年度/参数敏感性 | 是否稳健 | 默认风格：brief。
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-07 | pass→pass | 16,080 | 15,182 | -6% | 1 | 1 | 0% | 2,674 | 2,360 | -12% | 0 | 0 | — |
case-01 | fail→fail | 29,014 | 9,444 | -67% | 1 | 1 | 0% | 4,443 | 972 | -78% | 0 | 0 | — |
case-02 | fail→fail | 59,530 | 18,205 | -69% | 1 | 1 | 0% | 4,179 | 1,328 | -68% | 0 | 0 | — |
case-03 | fail→fail | 24,895 | 15,350 | -38% | 1 | 1 | 0% | 3,686 | 1,016 | -72% | 0 | 0 | — |
case-04 | fail→pass | 20,701 | 7,030 | -66% | 1 | 1 | 0% | 3,072 | 1,484 | -52% | 0 | 0 | — |
case-05 | fail→fail | 18,915 | 33,225 | +76% | 1 | 1 | 0% | 3,414 | 3,728 | +9% | 0 | 0 | — |
case-06 | pass→pass | 14,675 | 13,985 | -5% | 1 | 1 | 0% | 2,335 | 2,503 | +7% | 0 | 0 | — |
case-12 | pass→pass | 19,145 | 15,462 | -19% | 1 | 1 | 0% | 3,238 | 3,124 | -4% | 0 | 0 | — |
case-13 | pass→fail | 21,899 | 14,654 | -33% | 1 | 1 | 0% | 2,727 | 1,359 | -50% | 0 | 0 | — |
case-19 | pass→pass | 16,968 | 12,339 | -27% | 1 | 1 | 0% | 2,512 | 2,490 | -1% | 0 | 0 | — |
case-20 | fail→fail | 10,069 | 12,510 | +24% | 1 | 1 | 0% | 1,306 | 1,797 | +38% | 0 | 0 | — |
case-08 | pass→fail | 17,587 | 14,567 | -17% | 1 | 1 | 0% | 2,304 | 1,004 | -56% | 0 | 0 | — |
case-09 | fail→pass | 19,794 | 13,080 | -34% | 1 | 1 | 0% | 1,552 | 1,137 | -27% | 0 | 0 | — |
case-10 | fail→pass | 15,564 | 4,332 | -72% | 1 | 1 | 0% | 2,163 | 1,078 | -50% | 0 | 0 | — |
case-11 | fail→pass | 10,570 | 4,624 | -56% | 1 | 1 | 0% | 1,883 | 849 | -55% | 0 | 0 | — |
case-14 | fail→pass | 10,027 | 2,882 | -71% | 1 | 1 | 0% | 1,250 | 949 | -24% | 0 | 0 | — |
case-15 | pass→pass | 22,290 | 11,435 | -49% | 1 | 1 | 0% | 3,123 | 2,440 | -22% | 0 | 0 | — |
case-16 | pass→pass | 15,702 | 15,133 | -4% | 1 | 1 | 0% | 1,995 | 2,959 | +48% | 0 | 0 | — |
case-17 | fail→pass | 7,149 | 3,987 | -44% | 1 | 1 | 0% | 1,040 | 904 | -13% | 0 | 0 | — |
case-18 | fail→pass | 18,988 | 8,892 | -53% | 1 | 1 | 0% | 3,205 | 1,910 | -40% | 0 | 0 | — |
case-21 | fail→fail | 19,550 | 11,473 | -41% | 1 | 1 | 0% | 3,928 | 1,187 | -70% | 0 | 0 | — |
case-22 | fail→fail | 16,621 | 40,039 | +141% | 1 | 1 | 0% | 2,529 | 6,464 | +156% | 0 | 0 | — |
case-23 | pass→pass | 17,597 | 12,654 | -28% | 1 | 1 | 0% | 2,278 | 2,196 | -4% | 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 17 counted toward the lift figure. The other 6 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 +22 percentage points is the difference between those two pass rates over the 17 comparable cases. 4 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.