Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Open-source crypto trading bot. Strategy development in Python, backtesting, hyperparameter optimization, dry-run and live trading. Supports major exchanges via CCXT. Telegram integration for monitoring.
.claude/skills/mkurman-freqtrade/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-07 | ✗→✓ | ▲ Improved | -32% | 0% |
| case-08 | ✗→✓ | ▲ Improved | -42% | 0% |
| case-14 | ✗→✓ | ▲ Improved | -35% | 0% |
| case-18 | ✗→✓ | ▲ Improved | -44% | 0% |
| case-19 | ✗→✓ | ▲ Improved | -28% | 0% |
Freqtrade is an open-source crypto trading bot written in Python. Supports strategy development, backtesting, hyperparameter optimization, and dry-run or live trading via 100+ exchange backends (CCXT).
bashgit clone https://github.com/freqtrade/freqtrade.git cd freqtrade uv pip install -e .
pythonfrom freqtrade.strategy import IStrategy class MyStrategy(IStrategy): timeframe = "1h" minimal_roi = {"0": 0.01} stoploss = -0.05 def populate_indicators(self, dataframe, metadata): dataframe["rsi"] = 100 - (100 / (1 + dataframe["close"] / dataframe["close"].shift(14))) return dataframe def populate_buy_trend(self, dataframe, metadata): dataframe.loc[(dataframe["rsi"] < 30) & (dataframe["volume"] > 0), "buy"] = 1 return dataframe
bashfreqtrade backtesting --strategy MyStrategy --timerange 20240101-20241231 freqtrade trade --strategy MyStrategy --dry-run
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 12,551 | 10,987 | -12% | 1 | 1 | 0% | 2,481 | 2,442 | -2% | 0 | 0 | — |
case-02 | pass→pass | 7,675 | 5,287 | -31% | 1 | 1 | 0% | 1,515 | 1,333 | -12% | 0 | 0 | — |
case-03 | pass→pass | 8,699 | 3,284 | -62% | 1 | 1 | 0% | 1,450 | 820 | -43% | 0 | 0 | — |
case-04 | pass→pass | 7,475 | 1,589 | -79% | 1 | 1 | 0% | 1,278 | 573 | -55% | 0 | 0 | — |
case-05 | pass→pass | 4,405 | 2,107 | -52% | 1 | 1 | 0% | 850 | 611 | -28% | 0 | 0 | — |
case-06 | pass→pass | 13,786 | 6,289 | -54% | 1 | 1 | 0% | 2,523 | 1,585 | -37% | 0 | 0 | — |
case-07 | fail→pass | 17,269 | 8,821 | -49% | 1 | 1 | 0% | 2,820 | 1,924 | -32% | 0 | 0 | — |
case-08 | fail→pass | 12,550 | 5,540 | -56% | 1 | 1 | 0% | 2,150 | 1,249 | -42% | 0 | 0 | — |
case-09 | pass→pass | 6,964 | 3,106 | -55% | 1 | 1 | 0% | 1,215 | 870 | -28% | 0 | 0 | — |
case-10 | fail→fail | 7,489 | 3,807 | -49% | 1 | 1 | 0% | 1,211 | 986 | -19% | 0 | 0 | — |
case-11 | fail→fail | 4,445 | 3,918 | -12% | 1 | 1 | 0% | 765 | 897 | +17% | 0 | 0 | — |
case-12 | pass→pass | 8,625 | 2,581 | -70% | 1 | 1 | 0% | 1,339 | 714 | -47% | 0 | 0 | — |
case-13 | fail→fail | 13,787 | 12,610 | -9% | 1 | 1 | 0% | 2,870 | 2,968 | +3% | 0 | 0 | — |
case-14 | fail→pass | 5,543 | 2,079 | -62% | 1 | 1 | 0% | 1,089 | 704 | -35% | 0 | 0 | — |
case-15 | pass→pass | 2,563 | 1,524 | -41% | 1 | 1 | 0% | 395 | 528 | +34% | 0 | 0 | — |
case-16 | pass→pass | 10,230 | 4,159 | -59% | 1 | 1 | 0% | 2,043 | 1,140 | -44% | 0 | 0 | — |
case-17 | pass→pass | 10,899 | 2,607 | -76% | 1 | 1 | 0% | 1,942 | 743 | -62% | 0 | 0 | — |
case-18 | fail→pass | 6,457 | 1,981 | -69% | 1 | 1 | 0% | 1,034 | 583 | -44% | 0 | 0 | — |
case-19 | fail→pass | 5,161 | 1,956 | -62% | 1 | 1 | 0% | 902 | 649 | -28% | 0 | 0 | — |
case-20 | pass→pass | 7,620 | 6,447 | -15% | 1 | 1 | 0% | 1,525 | 1,657 | +9% | 0 | 0 | — |
case-21 | pass→pass | 11,247 | 7,532 | -33% | 1 | 1 | 0% | 2,171 | 1,740 | -20% | 0 | 0 | — |
case-22 | pass→pass | 18,543 | 16,150 | -13% | 1 | 1 | 0% | 3,301 | 3,263 | -1% | 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 +23 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.