3 things I changed in my AI agent strategy description that doubled the quality of its trades

Been using an AI agent on 1024EX beta for about 6 weeks now. First 2 weeks were meh. Then I realized the problem was my instructions, not the agent. Here's what I changed:

\*\*1. Added time filters\*\*

Before: "Trade BTC momentum when RSI crosses above 50 with volume confirmation"

After: "Trade BTC momentum when RSI crosses above 50 with volume confirmation. Only enter positions when current 4H volume exceeds 120% of the 20-period average for that specific time window. No entries between 23:00-05:00 UTC on Saturdays or Sundays."

Result: Eliminated garbage trades during dead hours. Win rate went from 52% to 61%.

\*\*2. Made the exit criteria dynamic\*\*

Before: "Stop loss 2%, take profit 5%"

After: "Adjust stop loss and take profit based on current ATR. Minimum stop loss 1%, maximum 3%. If ATR is elevated above 2x the 20-period average, tighten stops to 1.5% maximum."

Result: Agent started taking faster profits in volatile conditions and letting winners run in steady trends. Average loser shrank from -$35 to -$19.

\*\*3. Told it when NOT to trade (this was the biggest one)\*\*

Before: Nothing about when to stay out.

After: "Do not open new positions when BTC's 4H Bollinger Bandwidth is below \[threshold\] (compression phase). Wait for expansion before considering entries."

Result: Stopped getting chopped up during ranging markets. Week 3 losses dropped significantly.

\*\*The lesson:\*\* These agents are literal. They do exactly what you tell them. The more specific and detailed your instructions, the better the output. Treat it like writing a spec doc, not a wish.

hope this helps anyone else testing this stuff. feel free to ask questions.