monitor

Monitor - Autonomous Learning Loop

Closes the autonomous loop: check trades, close hits, evaluate predictions, generate summaries.

All agents use subagent_type: general-purpose with explicit model to ensure MCP tool access. Include "Do NOT use the Edit tool" in every prompt.

Workflow

Step 1: Get Current Prices

Delegate using Task with subagent_type: general-purpose, model: haiku:

"You are the market-monitor agent. Get current prices for ALL symbols that have open trades. Call get_portfolio_state() from crypto-learning-db first to see which symbols have open positions. Then use get_exchange_prices(symbol=...) from crypto-exchange MCP to get live prices for each symbol. Return a JSON object like: {"BTC/USDT": 98500, "ETH/USDT": 3200} Do NOT use the Edit tool."

Step 2: Check Open Trades Against SL/TP

Using the prices from Step 1 and the open trades from get_portfolio_state():

For each open trade, check:

• Long trade: Did price drop to or below stop_loss? Did price rise to or above take_profit?
• Short trade: Did price rise to or above stop_loss? Did price drop to or below take_profit?

If SL or TP was hit, delegate using Task with subagent_type: general-purpose, model: opus:

"You are the portfolio-manager agent. Close trade {trade_id}. Current price is ${price}. The {SL/TP} at ${level} was hit. Call close_trade(trade_id='{trade_id}', exit_price={price}, close_reason='{SL/TP} hit at ${level}') from crypto-learning-db. Do NOT use the Edit tool."

Step 2b: Trailing Stop Adjustment

For open trades that were NOT closed (still active), check if the trade is profitable:

• Long: current price > entry_price
• Short: current price < entry_price

If profitable AND the current stop_loss hasn't been optimally trailed, delegate using Task with subagent_type: general-purpose, model: sonnet:

"You are the risk-specialist agent. Read agents/risk-specialist.md for context. Trade {trade_id} ({symbol}, {side}) is profitable. Entry: ${entry}, Current: ${price}, SL: ${stop_loss}, TP: ${take_profit}. Analyze whether to trail the stop-loss. Use calculate_volatility(symbol=...) and get_support_resistance(symbol=...) from crypto-technical MCP. If you recommend adjusting, call update_trade(trade_id='{trade_id}', stop_loss={new_sl}, notes='your reasoning') from crypto-learning-db. Rules: only trail in profitable direction, never widen the stop, leave room for normal volatility. Do NOT use the Edit tool."