How I automated my gold trading analysis (step by step)

I trade XAUUSD (gold against the US dollar) on ThinkMarkets, using MT5 on mobile. My session is 18:00-19:00 SAST — the US session open, when gold moves the most.

Before automation, I did everything manually: checked the price, looked up S/R levels, scanned the news, calculated position size, and logged trades after the session. It took 20-30 minutes of preparation and another 10 minutes of post-session logging.

Now the agent handles the prep, the logging, and the risk enforcement. Here's how it was built — one piece at a time.

---

Step 1: The trading profile

Before automating anything, I documented the entire trading approach in a profile:

• Instrument: XAUUSD only (gold is what I know)
• Session: 18:00-19:00 SAST (US open)
• Strategy: Support/resistance levels + momentum continuation
• Timeframes: H4 for bias → M15 for entry → M5 for precision
• Risk: 1-2% per trade, max 2 trades per session, 3% daily loss limit
• Targets: TP1 at 1:1.5 R&R, TP2 at 2:1
• Goals: >55% win rate, ≥1.5:1 average R&R, 3-8% monthly ROI

This profile became the foundation for everything that followed. You can't automate what you haven't defined.

---

Step 2: Morning briefing (Day 1)

The first automated tool was a morning briefing that runs daily at 15:00 SAST — before the trading session opens.

It fetches:

• Live gold price from gold-api.com (with fallback to metals.live and investing.com)
• Key support and resistance levels
• Economic calendar events that could affect gold
• A pre-session checklist (what to watch, what to avoid)

From zero to a working morning briefing on Day 1. It was the first cron job I ever deployed.

---

Step 3: Evening recap (Day 1)

Built on the same day — an evening recap at 20:00 SAST, right after the session closes.

It summarizes:

• What happened during the US session
• How price interacted with the key levels
• What the setup was (or wasn't)
• Outlook for the next day

This created a feedback loop: morning briefing sets the plan, evening recap reviews what happened. But there was still a gap — no way to log the actual trades.

---

Step 4: Trade journal (Night 3 — built by the nightly ideas agent)

The agent noticed we had analysis automation but no trade logging. It built a complete trade journal:

• Add trades with instrument, direction, entry, stop-loss, take-profit, result
• Stats calculation: win rate, average R&R, total and daily P&L, streak, max drawdown
• Data storage in JSON at ~/.hermes/xauusd_journal/trades.json
• Daily prompt at 19:15 — right after the session — asking me to log the day's trades

This created the missing feedback loop. Now I can see if my actual results match my targets.

---

Step 5: Session prep pinger (Night 4 — built by the nightly ideas agent)

The morning briefing looks at yesterday's data. I needed something that runs right before the session with live, current data.

The session prep pinger runs at 17:45 — 15 minutes before 18:00. It:

• Fetches live gold price from 3 sources
• Loads the day's S/R levels from the morning briefing's output
• Checks the economic calendar for any events in the next hour
• Outputs a concise checklist: bias, key levels, what to watch, what to avoid

First test run: fetched gold at $4,541.20, loaded levels correctly, output clean checklist. Working.

---

Step 6: Risk calculator (Night 5 — built by the nightly ideas agent)

Having risk rules documented is one thing. Having a tool that enforces them before you enter a trade is another.

The risk calculator:

• Takes account balance, risk percentage, and stop-loss distance
• Calculates position size in ounces
• Enforces the 3% daily loss limit
• Enforces max 2 trades per session
• Enforces minimum 1:5:1 R&R
• Reads S/R levels from the morning briefing's daily output
• Auto-resets daily state at midnight

If a trade doesn't meet the criteria, it tells you why. Before you enter, not after.

---

What the full workflow looks like now

15:00 — Morning briefing arrives in Discord. I review the plan.

17:45 — Session prep pinger arrives. Live price, S/R levels, checklist. I'm ready.

18:00-19:00 — I trade. The agent doesn't trade for me — I make the decisions. But I'm working from a prepared plan, not doing analysis on the fly.

19:15 — Trade journal prompt arrives. I log the session's trades.

20:00 — Morning recap arrives. Reviews what happened vs what was planned.

Next morning — I have data. Win rate, R&R, P&L, streaks. I can see if I'm hitting my targets.

---

What's still manual

I still make the actual trading decisions. The agent prepares, calculates, and logs — but doesn't enter or manage trades. That's a deliberate choice. The discretionary judgment of when to enter, when to wait, and when to walk away is the part I trust myself with.

The next phase (Phase 2, months 2-4) includes semi-automated signal detection — S/R scanners, momentum candle detection, and breakout alerts. But live automated execution is Phase 4 (MQL5 expert advisor), and that's a long way off.

---

The real lesson

Automating trading isn't about replacing the trader. It's about removing everything around the decision that isn't the decision itself. Analysis, preparation, risk calculation, logging, performance tracking — all of that can and should be automated.

What remains is the part that requires human judgment: reading the price action, managing the trade, knowing when the plan has changed.

That's the part worth spending your mental energy on.

---

Trading profile, strategy, and automation tools documented in the Obsidian vault. Win rate and performance targets are tracked in the automated trade journal. Nothing in this post is hypothetical — every tool described here is running on the live system.