strategy-compare — Claude Code skill: side-by-side strategy stats

What strategy-compare does

The skill takes a symbol plus a list of strategies and writes a script to backtesting/strategy_comparison/ named {symbol}_strategy_comparison.py. Data is fetched once through OpenAlgo, every strategy runs on the same series, and the metrics land in one table: total return, Sharpe, Sortino, max drawdown, win rate, trade count and profit factor. A NIFTY benchmark is added via NSE_INDEX, and equity curves for all variants are overlaid in Plotly with template="plotly_dark". The comparison is saved to CSV.

Indicators come from OpenAlgo ta by default, not from VectorBT built-ins. Switching to TA-Lib happens only when you ask for it by name. For Supertrend and Donchian there is no choice: TA-Lib has no equivalent, so OpenAlgo ta handles them. Signals are cleaned with ta.exrem(), and .fillna(False) always runs before that call.

Pass a DuckDB path and the script reads straight from it with duckdb.connect(path, read_only=True). In that standalone mode openalgo.ta may not import, so an inline exrem() fallback is used instead. Fees are set for Indian equity delivery: fees=0.00111 and fixed_fees=20. With no strategies listed, the skill compares ema-crossover, rsi, donchian and supertrend; the token long-vs-short switches the run to longonly, shortonly and both for the first real strategy.