What problem does it solve? Manually designing, testing, and validating quantitative alpha factors is slow and error-prone. This Skill orchestrates the full RD-Agent pipeline—environment preflight, LLM-driven factor evolution, batch backtesting, IC/Sharpe ranking, LLM explanation, and production export—so a single command produces validated factors and a Markdown report. ## Core Features & Use Cases - One-command pipeline: Runs preflight health checks (conda shim, litellm patch, DeepSeek key), launches evolution, polls task status, batch-backtests factors, ranks by |IC|, explains top factors, and exports qualifying factors to the production feature library. - Factor factory mode: Systematically derives thousands of expression factors from QuantDB fields using operators, windows, and binary combinations, computing IC/ICIR with correlation deduplication. - Manual API control: Provides curl-based endpoints for evolve, polling, cancel, backtest, explain, export, and stats when fine-grained control is needed. - Use Case: A quant researcher wants to test the hypothesis "chip concentration rising with low-position turnover expansion" on the CSI 300. They run the pipeline with that direction, wait 30–90 minutes, and receive a ranked factor list with IC/Sharpe metrics, LLM interpretations, and a Markdown report. ## Quick Start Ask the assistant to run the RD-Agent factor mining pipeline with a direction like "连板高度递减与涨停回封率" on the csi300 universe and generate the factor report.
