Strategy Skill

Strategy Skill

Purpose

Take a structured strategy configuration — entry/exit conditions, TP/SL, drawdown limits, seed capital — and produce a runnable JavaScript strategy script that returns {trades, equity}. After the script is backtested, the same skill can analyze the results and suggest concrete improvements.

This skill is the counterpart to the Pattern Skill: instead of finding historical patterns, it trades them (or any other hypothesis).

When to use this skill

Vibe Trade should dispatch to the Strategy Skill when the user:

• Wants to build a new strategy from scratch — activate chatbox.card.strategy_builder to show the structured form in the chat
• Has submitted the Strategy Builder form — generate the JS from context.strategy_config
• Asks to analyze backtest results — use context.analyze_results plus the strategy config to produce an assessment + suggestions
• Wants to iterate on an existing strategy (tweak TP/SL, change the entry rule) — regenerate from the updated config

Instructions

1. If no strategy_config is in context, show the structured form:

   • Emit chatbox.card.strategy_builder to inject the form into the chat
   • Respond with a brief prompt like "Fill in the fields and I'll generate a runnable strategy for you."

2. If strategy_config is present and mode = "generate" (default):

   • Generate a JS strategy script that:
     • Defines every indicator helper it uses (don't assume sma() exists)
     • Handles edge cases (bounds check, initial lookback period)
     • Tracks maxAdverseExcursion / maxFavorableExcursion per trade
     • Pushes to the equity array on every bar (not just on trade events)
     • Respects the max drawdown limit
   • Emit script_editor.load with the full script
   • Emit bottom_panel.activate_tab → portfolio

3. If mode = "analyze" (backtest results in context.analyze_results):

   • Produce a 2–3 sentence overall assessment (profitability + risk)
   • Call out specific strengths and weaknesses in the metrics
   • Return 3–5 concrete improvement suggestions as data.suggestions
   • Do NOT regenerate the script — only analysis in this mode

4. Never touch the script generation without a config. Free-form strategy chat (without a structured config) should be handled as a general chat fallback, not by this skill.

Inputs

strategy_config shape:

{
  entryCondition: string,
  exitCondition: string,
  takeProfit: { type: "percentage" | "absolute", value: number },
  stopLoss: { type: "percentage" | "absolute", value: number },
  maxDrawdown: number,    // percent
  seedAmount: number,     // USD
  specialInstructions: string
}

Outputs

Generate mode:

• reply — short confirmation ("Strategy script generated.")
• script — JavaScript strategy returning {trades, equity}
• script_type — "strategy"
• data.config — echoes the config (for persistence)
• tool_calls — load + activate portfolio tab

Analyze mode:

• reply — 2–3 sentence assessment
• data.suggestions — 3–5 improvement suggestions
• tool_calls — none (analyze is read-only)

Tools used

Examples

Generate Config: Entry = SMA(20) > SMA(50), Exit = SMA(20) < SMA(50), TP = 5%, SL = 2%, Max DD = 20%, Seed = $10,000.

→ Returns a strategy script with inline sma() helper, golden-cross entry logic, per-trade PnL + MAE/MFE tracking, equity-curve pushes on every bar. Loads into the editor; bottom panel switches to Portfolio.

Analyze Metrics: 42 trades, 38% win rate, 1.1 profit factor, -18% max drawdown.

→ "Marginal profitability, asymmetric losses are hurting overall returns." Suggestions: tighten stop to 1.5%, add an ATR volatility filter, require ADX > 25 for trend regime, etc.

Underlying implementation

Wired through core/agents/processors.py::_strategy_processor, which calls core/agents/strategy_agent.py::StrategyAgent.generate_from_config or .analyze_results depending on context.mode. The agent owns STRATEGY_GENERATE_PROMPT and STRATEGY_ANALYSIS_PROMPT.