TraderAlice/Auto-Quant — reverse-engineered prompt

Build me a local Auto Quant research sandbox where an AI coding agent can safely test crypto trading strategy ideas without touching real money.

Use FreqTrade backtesting in dry run mode, with a fixed config for BTC, ETH, SOL, BNB, and AVAX across 1h, 4h, and 1d data. Include a setup script that downloads the historical Binance data, and a run script that finds every strategy in a strategies folder, runs the backtests, and prints a clear summary with Sharpe, max drawdown, profit, and per pair results.

The main thing I want is the autonomous research loop. Put the instructions in a program.md file so an agent can create up to 3 strategies, evolve them, fork them, kill weak ones, rerun tests, and log each decision to results.tsv. It should be able to commit good changes and reset bad ones.

Also include a simple analysis notebook or chart so I can review the experiment history afterward. Make it clear this is research only, not trading advice.