For Day 18, I’ve created a new flowchart for my trading agent system. It helped me clarify how the components should work together—and where human approval belongs.
As a Windows sysadmin and a beginner in AI programming, I’m combining familiar infrastructure with something new: a trading workflow built around MetaTrader 5, Python, and AI-assisted analysis.
Here’s the plan:
📊 Strategies and market data
Multiple strategy modules, with price data feeding the strategies and an economic calendar informing the risk checks.
🛡️ Risk checks before every order
Every proposed trade must pass through the risk module. Hard risk limits will be enforced in code, outside the AI’s authority to modify. Only I can deliberately change those limits.
🧠 AI suggests; I approve
Claude Code will help analyse logs—including rejected trades and their reasons—and propose improvements. Approved changes must then go through backtesting and demo testing before being considered for live use.
📋 Visibility and monitoring
Daily Excel and Word reports, urgent MT5 notifications, and heartbeat monitoring to flag outages. Private GitHub version control will help me track code changes.
💶 Money stays under manual control
The agent will have no responsibility for deposits or withdrawals. Those remain entirely with me.
The first target is a working demo setup. Live trading and optional integrations come later, after testing and a deliberate decision.
This is still an architecture plan, not a finished system. Drawing it out helped me separate what I want to build now from what can wait.
What would you simplify or clarify before moving from this blueprint to implementation? 👇