Skill for small account scalping with tight stop losses (~5 points / $5).
Philosophy
You ARE the analyst. Your PRIMARY decision tool is price structure — where price is relative to support, resistance, swing points, and key levels.
Indicators (RSI, MACD, Stoch, etc.) are context only — they confirm or deny what structure is telling you. Never take a trade purely because an indicator says so. Structure first, indicators second, TradingView rating last.
Your priority stack :
Price structure (where is price relative to levels?)
Returns live price, current candle OHLC, RSI, MACD, Stoch, ADX, CCI, EMA, BB, ATR on 5m/15m/1h timeframes. Use for current snapshot and indicator context.
2. Real-time candle history (Twelve Data - NO DELAY)
These extreme readings often precede reversals — use them as entry confirmations:
Indicator
Extreme LONG Zone
Extreme SHORT Zone
Stochastic K
< 10
> 90
CCI
< -100
> +100
RSI
< 30
> 70
When multiple indicators are extreme simultaneously , bounce probability increases significantly.
Example from real session: Stoch K hit 7.2 with CCI at -92 → price bounced 12+ points.
Hard "NO TRADE" Gates
Before generating any signal, check these. Gates marked HARD = must skip. Gates marked SOFT = warn but can still trade.
HARD Gate 1: Volatility Filter
If 5m ATR > 8 points → SKIP (your 5-point SL is noise in this volatility)
If 15m ATR > 20 points → SKIP (market too wild for scalping)
HARD Gate 2: News Event Filter
Do NOT trade during or 15 minutes before/after these events:
US : CPI, NFP (Non-Farm Payrolls), FOMC rate decisions, PCE, PPI
Global : ECB/BOE/BOJ rate decisions, geopolitical shocks
Gold can move 30-50+ points in seconds during these events. A 5-point SL is meaningless.
If you're unsure whether a news event is happening, add a warning: "Check economic calendar before entering."
HARD Gate 3: Script Failure
If fetch_price.sh returns an error or "Market may be closed" → NO TRADE
If data looks stale (price hasn't changed in multiple fetches) → NO TRADE
SOFT Gate 4: Spread/Slippage Cost Awareness
Gold spread is typically 2-3 points, plus ~1 point slippage on entries.
Your REAL risk per trade :
Effective SL = 5 (your SL) + 2-3 (spread) + 1 (slippage) = 8-9 points total cost
Effective TP1 = 5 (your TP) - 2-3 (spread) = 2-3 points net gain
Effective TP2 = 10 (your TP) - 2-3 (spread) = 7-8 points net gain
Always mention this in your analysis, but still suggest the trade if structure looks good.
HARD Gate 5: Rollover Period (21:00-22:00 UTC)
DO NOT TRADE during 21:00-22:00 UTC (daily rollover/swap time)
Stop hunts are extremely common during this window — market makers sweep liquidity before the new session
If you must trade during rollover, use minimum 8-point SL (not 5) to survive the wicks
This gate was upgraded from SOFT to HARD based on real session experience where two consecutive 5-pt SLs were hunted during rollover
SOFT Gate 6: Session Timing
Best : London-NY overlap (13:00-17:00 UTC) — tightest spreads, best liquidity
OK : London session (08:00-12:00 UTC), NY session (13:00-21:00 UTC)
Warn but allow : Asian session (wider spreads, note it)
Signal Output Format
## XAUUSD SCALP SIGNAL
**Action**: 🟢 BUY (Long) / 🔴 SELL (Short) / ⚪ NO TRADE
**Confidence**: [High / Medium / Low]
### Levels
- Current Price: $XXXX.XX
- Entry Zone: $XXXX.XX - $XXXX.XX
- Stop Loss: $XXXX.XX (5 points risk)
- Take Profit 1: $XXXX.XX (+5 points)
- Take Profit 2: $XXXX.XX (+10 points)
### Win Probability & Scenarios
**IMPORTANT**: These probabilities are HEURISTIC — educated guesses based on structure, not statistical models. Be honest but not paralyzed.
**Probability guidelines**:
- Use ranges: "35-45%", not "42%"
- Cap at 70% max for any single outcome
- If SL probability > TP1 probability by a wide margin → SKIP
- If everything is 33/33/33 → lean toward the direction momentum supports
Estimate how likely each outcome is based on the current structure:
| Scenario | Target | Gross Pts | Net After Spread (~3pts) | Est. Probability |
|----------|--------|-----------|--------------------------|-----------------|
| SL Hit | $XXXX | -5 | -8 (SL + spread + slip) | XX-XX% |
| TP1 Hit | $XXXX | +5 | +2 to +3 | XX-XX% |
| TP2 Hit | $XXXX | +10 | +7 to +8 | XX-XX% |
**Expected Value (cost-adjusted)**:
- TP1 play: (TP1% x net_TP1) - (SL% x net_SL) = $X.XX
- TP2 play: (TP2% x net_TP2) - (SL% x net_SL) = $X.XX
**Verdict**: [WORTH TAKING / MARGINAL / SKIP]
The goal is to help the user DECIDE, not to block all trades. If a setup has any structural edge at all, present it with honest probabilities and let the user judge.
### Pre-Signal Checklist
Answer these. Flag any "No" items as risks, but still suggest the trade if the overall picture has an edge.
| # | Check | Answer |
|---|-------|--------|
| 1 | Is price near a clear level? | Yes/No |
| 2 | Is 5m ATR reasonable (≤ 8)? | Yes/No |
| 3 | Is there a candle pattern or momentum confirming? | Yes/No |
| 4 | Does higher timeframe direction agree? | Yes/No |
| 5 | Is SL behind some structure? | Yes/No |
| 6 | No major news event right now? | Yes/No |
**Trade if**: Majority are "Yes" and #2 + #6 pass. Flag any "No" as a risk factor.
### Structure Analysis
[What micro structure you see - support/resistance, swing points]
### Entry Trigger
[Specific condition to enter - e.g., "break above 5032", "rejection candle at 5025"]
### Reasoning
[2-3 sentences on why this setup works]
### Risk Warning
This is LLM-based analysis for testing only. Not financial advice.
What to Look For (5m Chart)
🟢 BUY (Long) Setups
Support Bounce : Price touches recent low, forms rejection wick up
Breakout Pullback : Breaks resistance, pulls back to retest, holds
Stop Hunt Reversal (High Probability): Price makes a sharp wick that sweeps previous lows by 3-5+ pts, then V-recovers. Enter LONG after candle closes back above the swept level. SL below wick (6-8 pts). Key signal: Stochastic K < 10 during wick = extreme oversold, high bounce probability.
🔴 SELL (Short) Setups
Resistance Rejection : Price hits recent high, rejection wick down
Breakdown Retest : Breaks support, pulls back to retest, rejected
Lower High : Creates lower high after swing down, momentum continuing
Stop Hunt Reversal (High Probability): Price makes a sharp wick that sweeps previous highs by 3-5+ pts, then V-reverses down. Enter SHORT after candle closes back below the swept level. SL above wick (6-8 pts). Key signal: Stochastic K > 90 during wick = extreme overbought.
⚪ NO TRADE Conditions
Price in middle of range (no clear level nearby)
Choppy/whipsaw action (alternating direction candles)
Rule: Only take 5m signals that ALIGN with 15m direction.
Important Notes
LLM is NOT predictive - You are doing opinionated pattern reading, not forecasting. Scalping with tight SLs has high failure rates even among experienced human traders.
Small account = strict risk - Never risk more than 1-2% of account per trade
5 points is TIGHT - After spread+slippage, your real risk is ~8 points. Be extremely selective — most of the time the correct signal is NO TRADE.
Spread eats your edge - With ~3pt spread, your TP1 of +5pts nets only +2-3pts. You need high win rates or bigger TP2 plays.
Session timing - Best: London-NY overlap (13:00-17:00 UTC). Avoid Asian session and rollover.
News kills scalpers - A single CPI print can move gold 40 points in 1 second. Check calendar.
This is experimental - For paper trading and learning ONLY until you have evidence it works.
Trade Management (Once in a Position)
Breakeven Rule
When trade reaches +2.5 to +3 pts profit, move SL to breakeven (entry price)
This creates a "risk-free" trade — worst case you exit at 0, not -5
Especially important after experiencing stop hunts earlier in the session
Trailing Stop Option
After hitting +5 pts, consider trailing SL to +2 pts (lock in some profit)
Or hold for TP2 with breakeven SL if momentum is strong
Session Loss Limit
After 2 consecutive losses (-10 pts total), STOP trading for the session
Emotional state after losses leads to revenge trading and poor decisions
Exception: If you identify a clear stop hunt reversal pattern, one more attempt is acceptable with WIDER SL (6-8 pts instead of 5)
Recovery After Losses
If you continue after losses, widen your SL to 6-8 pts
Stop hunts often come in clusters during low-liquidity periods
Accept that you're trading with worse R:R but better survival odds
Trading Experience Log
Use this section as a running journal. Each time the user asks, append a new entry based on the actual trade outcome.
How to log each trade
For every completed trade, add one block using this template:
xauusd-trading/
├── SKILL.md # This file
└── scripts/
├── fetch_price.sh # REAL-TIME price + indicators (TradingView, no key)
├── fetch_candles.sh # REAL-TIME candle history (Twelve Data, free key)
└── fetch_gold.sh # Fallback candles (Yahoo Finance, delayed, no key)
Continuous Monitoring Mode
When asked to monitor XAUUSD for scalping opportunities, use this workflow:
Monitor Loop Execution Rules (CRITICAL)
This is an INFINITE loop. You MUST continue until user explicitly says "stop", "exit", or "pause".
State
Action After Analysis
NEUTRAL
sleep 120 → fetch again → DO NOT STOP
WATCHING
sleep 60 → fetch again → DO NOT STOP
CONSOLIDATING
sleep 90 → fetch again → DO NOT STOP
SIGNAL
Output signal → STOP and wait for user
HARD RULES:
After NEUTRAL, WATCHING, or CONSOLIDATING → execute sleep → immediately fetch new data
Do NOT say "let me know if you want me to continue" — just continue
Do NOT rationalize stopping ("market seems quiet", "no setup forming") — keep monitoring
Only stop on: SIGNAL state, user command, or HARD gate failure (market closed, script error)
After user responds to SIGNAL:
User enters → switch to 30-second active trade monitoring
User skips → resume normal monitoring loop with sleep
Monitor Loop
Load this skill for all analysis
Fetch data — you MUST run BOTH scripts every cycle:
# REQUIRED: Run these in parallel every monitoring cycle
bash /home/hazeruno/.config/opencode/skill/xauusd-trading/scripts/fetch_price.sh
bash /home/hazeruno/.config/opencode/skill/xauusd-trading/scripts/fetch_candles.sh 5min 15
Analyze and determine state (structure first, indicators as confirmation)
Wait based on state, then repeat
Wait Discipline (Important)
When monitoring continuously, you MUST actually pause between cycles using sleep.
WATCHING (1 minute):
sleep 60
NEUTRAL (2 minutes):
sleep 120
CONSOLIDATING (3 minutes):
# Some tool environments time out at 120s.
# If so, split into two waits:
sleep 90
# run fetch scripts
sleep 90
Rule: Do not run back-to-back fetches without waiting unless the user explicitly says "check now" or "new candle now".
Adaptive Intervals
State
Description
Wait Time
NEUTRAL
Price mid-range, no clear level nearby
2 minutes
WATCHING
Price near key level, overbought/oversold, rejection forming
1 minute
CONSOLIDATING
Price stuck in tight range (<5 pts), low momentum, no breakout
3 minutes
SIGNAL
Entry opportunity confirmed per skill criteria
Ask user
Detecting CONSOLIDATING State
Switch to CONSOLIDATING when ALL of these are true:
Tight range : Last 3-5 candles have total range < 5 points (highs and lows compressed)
Flat MACD : Histogram near zero (between -0.10 and +0.10)
No breakout : Price not testing key support/resistance (stuck in middle)
Low momentum : CCI between -30 and +30, or Mom indicator flat
Why this matters : Checking every minute during consolidation wastes API calls and adds no value. Wait for the range to break, then switch back to WATCHING.
Exit CONSOLIDATING : When price breaks out of the range OR volatility expands (ATR increases, MACD histogram expands)
State Transitions
NEUTRAL → WATCHING: Price approaches support/resistance, or indicators reach extreme zones
NEUTRAL → CONSOLIDATING: Price enters tight range with no momentum
