菜单
Use this when the operator asks whether a factor / signal predicts returns — e.g. "does momentum work", "is low-vol still an edge", "rank these factors", "what's working in the market right now". Drives an evaluate → read-IC → context → honest verdict loop over the Alpha Zoo.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Use this whenever the operator wants to test a trading idea on history — e.g. "backtest a 20/50 moving-average cross on SPY", "does RSI mean-reversion work on AAPL", "how would a breakout strategy have done on BTC". Drives a specify → backtest → benchmark → check-robustness → honest verdict loop.
Use this ONLY when the operator explicitly asks to place, size, or manage a simulated/paper order — e.g. "buy 10 shares of AAPL in the paper account", "paper-trade this setup", "check my paper positions", "what's my paper P&L". Drives a confirm-armed → size → preview → operator-approval → fill loop. Never initiates a trade on your own.
Use this whenever the operator asks you to research, analyze, or give a read on a specific stock, ETF, or crypto — e.g. "what's the setup on NVDA", "should I be watching SPY", "analyze BTC", "bull/bear case for AAPL". Drives a quote → trend → fundamentals/context → news → structured read with risks.
Use this when the operator shares a trade journal / broker export and wants to understand their trading behavior — e.g. "review my trades", "what am I doing wrong", "analyze my journal", "why am I losing money even with a high win rate". Drives an ingest → profile → name-the-biases → Shadow-Account → fixes loop.
| name | evaluate-a-factor |
| description | Use this when the operator asks whether a factor / signal predicts returns — e.g. "does momentum work", "is low-vol still an edge", "rank these factors", "what's working in the market right now". Drives an evaluate → read-IC → context → honest verdict loop over the Alpha Zoo. |
| tools | ["factor_eval","factor_zoo","stock_price_history","web_search","memory_recall","current_time"] |
Test whether a factor actually predicts cross-sectional returns — by its Information Coefficient, not by story.
One factor → factor_eval(factor, universe, period). "What's working / rank
them" → factor_zoo(universe, period). Use the operator's universe if they have
one (a watchlist, a sector); else the default large-cap set. State the universe +
period — IC is sample-specific.
A factor's IC is regime-dependent. When low-vol "reverses," it usually means a
high-beta/risk-on regime (e.g. an AI mega-cap melt-up). Use web_search to name
the regime, and say it: "low-vol reversed because high-beta led this period."
Don't present a sample IC as a timeless law.
Lead with the factor's status (alive / weak / reversed / dead) + the number, then why (regime), then the caveat: this is one universe + one window; a factor alive here can be dead next year. Analysis, not advice — and definitely not a guarantee the factor keeps working.
