Prompt Engineering & Strategy Design

Understanding what data the AI has access to — and how to direct its analysis — is the key to getting the most value from SkyAnalyst AI. Whether you're writing custom automation prompts or asking questions in chat, this section shows you how to think about prompt design.

What Data the AI Receives

Every time you ask a question or an automation runs, the AI receives a comprehensive data package. You don't need to provide this data — it's gathered automatically. Understanding what's available helps you write better prompts.

Technical Indicators (3 Timeframes)

The AI receives 10 core indicators calculated across 5-minute, 15-minute, and 60-minute candles:

Key Insight: You don't need to describe these indicators in your prompts. The AI already sees them. Instead, tell the AI which indicators to prioritize or which conditions to look for.

Session Data

• Tokyo (9 AM - 3 PM JST): High and low prices during the Asian session
• London (8:30 AM - 4:30 PM GMT): High and low prices during the London session
• New York (9:30 AM - 4 PM EST): High and low prices during the NY session
• Active session status (which sessions are currently open)

Session levels are critical reference points. London highs/lows often define the range that NY either continues or reverses.

Market Context

The AI also receives cross-asset data:

• VIX: Volatility index (fear gauge)
• DXY: US Dollar Index
• 10Y Treasury Yield: Interest rate backdrop
• Oil: Commodity trend
• NYAD: NYSE Advance/Decline (market breadth)

These tell the AI about the broader environment. Rising VIX + falling NYAD = risk-off environment. Falling yields + weakening DXY = typically bullish for tech/gold.

Agent Analysis

When available, the AI also receives:

• Macro Agent output: Group bias, confidence, key factors, catalysts, risk events for your instrument's group
• Trend Agent output: Direction, confidence, strength, regime, key levels, invalidation for your specific instrument

See the AI Intelligence Agents guide section for details on how these agents work.

Dynamic Weighting

The AI dynamically adjusts how much weight it gives to technical vs fundamental analysis based on market conditions:

Pro Tip: During high-volatility events (VIX > 24), your prompts should emphasize macro factors and risk events. During calm markets, focus on technical levels and patterns.

Prompt Engineering Best Practices

Be Specific About What You Want

Instead of: "Analyze US30" Write: "Analyze US30 for potential long setups during the NY AM session, focusing on pullbacks to VWAP or EMA support with RSI confirmation"

Instead of: "Give me a trade" Write: "Scan NAS100 for breakout setups above the London session high with volume confirmation and at least 2:1 R:R"

The AI has all the data — your job is to tell it what to look for and what your conditions are.

Reference What the AI Already Knows

Good prompts leverage the data the AI already has:

• "Check if the Macro Agent bias aligns with this setup"
• "Factor in the Trend Agent's direction and confidence"
• "Assess whether the current VIX regime supports this trade type"
• "Compare the 15-minute and 60-minute EMA alignment"

Bad prompts re-describe what the AI can see:

• "The RSI is at 45 and MACD is crossing..." (the AI already has this data)
• "VIX is at 18 today..." (the AI already knows this)

Use Numbered Steps for Complex Analyses

Structure your prompt as a sequence of analytical steps:

Analyze EURUSD for NY AM session setups:
1. Check the economic calendar for high-impact USD/EUR events
2. Assess the Macro Agent bias and DXY direction
3. Evaluate 60-minute trend direction using EMA alignment
4. Find 5-minute entry zones at key levels (VWAP, Fibonacci, session levels)
5. Only present setups with at least 3 confluences

Numbered steps force the AI to work methodically rather than jumping to conclusions.

Set Confidence Gates

Tell the AI your quality standards:

• "Only present setups with HIGH probability (4+ confluences)"
• "If confidence is below Medium, state 'no setup' rather than forcing one"
• "Require both the Macro and Trend agents to agree before recommending full size"

This prevents the AI from presenting marginal setups just to give you an answer.

Tell the AI What to Prioritize

Different instruments respond to different drivers:

• Gold (XAUUSD): "Prioritize 10Y yield direction and DXY as primary filters, with technical entry on VWAP/Fibonacci"
• NAS100: "Prioritize yield sensitivity — this is the most rate-sensitive index"
• US30: "Prioritize NYAD breadth as the primary directional filter"
• Forex pairs: "Prioritize yield differentials and DXY direction as dominant drivers"

Common Mistakes to Avoid

Learning From Master Automation Prompts

The SkyAnalyst team maintains 6 master automation prompts (one per instrument). While you can't edit these, studying their structure reveals patterns you can apply to your own prompts.

The 7-Step Framework

Every master prompt follows a consistent structure:

1. Calendar Gate: Check economic events. If high-impact data is imminent, halt. If data just released, assess impact.
2. Macro Regime Assessment: Read the Macro Agent bias, check cross-asset signals (10Y yields, DXY, VIX, NYAD). Classify the fundamental backdrop.
3. Agent Synthesis: Read both the Macro and Trend Agents. When both agree with high confidence, maximum conviction. When they disagree, rules specify which agent to favor for that instrument.
4. Structural Framework: Identify key levels from daily reference data, session highs/lows, S/R zones.
5. Multi-Timeframe Analysis: 60-minute for bias, 15-minute for intermediate zones, 5-minute for entry precision.
6. Setup Construction: Define entry zone, trigger, stop loss (structural, scaled to volatility), and 3 take-profit levels (TP1 at 1R structural, TP2 at next level, TP3 at daily extreme).
7. Confidence Gate: Require minimum confluences (typically 4-5 of 6-8 criteria). List what's present and what's missing.

Key Patterns Across All Masters

Instrument-Specific Drivers:

• XAUUSD prompts prioritize 10Y yield + DXY as the "dominant driver pair"
• NAS100 prompts call 10Y yields "the most rate-sensitive index signal"
• US30 prompts use NYAD breadth as "the primary directional filter"
• EURUSD prompts treat DXY as "the dominant driver" with a hard rule against counter-DXY trades when Macro confidence > 70

Agent Hierarchy Rules:

• When agents agree: highest conviction
• When agents disagree on Gold/Forex: favor Macro Agent (these are fundamentally driven)
• When agents disagree on US Indexes: favor Trend Agent but reduce conviction

Stop Loss Philosophy:

• Always structural (beyond a real price level)
• Scaled to current volatility (wider stops in volatile markets, tighter in calm)
• If R:R becomes unworkable at structural stops, the answer is "No Trade" — not tighter stops

Confidence Scoring:

• Most masters require 4+ of 6-8 criteria
• At least 1-2 criteria must be "starred" (macro alignment or yield direction)
• Scores map to conviction: 4/7 = Medium, 5/7 = Medium-High, 6/7 = High, 7/7 = Very High

Building Your Own Strategy Prompts

Template: Macro-First Instrument Prompt

Use this for macro-sensitive instruments (Gold, Forex):

Analyze [INSTRUMENT] for [SESSION] session trade setups:

1. Economic calendar gate: Check for high-impact [CURRENCY] events within
   30 minutes. If pending, flag "WAIT." If released, note impact.

2. Macro regime: Read the Macro Agent bias and confidence. Check 10Y yield
   direction, DXY trend, and VIX level. Classify: bullish/bearish/mixed.

3. Agent synthesis: Compare Macro and Trend Agent signals. When both agree
   with confidence above 60, this is high conviction. When they disagree,
   favor the Macro Agent and cap confidence at Medium.

4. Multi-timeframe analysis: 60min for bias, 15min for zones, 5min for
   entry precision. Look for [YOUR PREFERRED PATTERNS].

5. Confluence gate: Require at least 4 of these: Macro Agent aligned,
   Trend Agent aligned, yield direction supports trade, EMA alignment
   confirms, price at key level (VWAP/Fibonacci/session), no imminent
   risk events. Only present qualifying setups.

Template: Technical-First Scalping Prompt

Use this for technically-driven instruments (US Indexes) in calm markets:

Scan [INSTRUMENT] for intraday scalp setups during the NY session:

1. Confirm VIX regime is Normal or below (if Elevated+, switch to
   wider-stop swing approach).

2. Identify the 5-minute trend using EMA 20 alignment and VWAP position.

3. Find pullback entries to EMA 20 or VWAP with RSI bouncing from
   40-50 zone (longs) or 50-60 zone (shorts).

4. Stop loss: 1x 5-minute ATR beyond entry level.

5. Targets: TP1 at 1R (nearest structure), TP2 at session high/low.

6. Reject if Trend Agent shows CHOPPY or RANGE_BOUND regime — these
   conditions favor mean-reversion, not momentum.

Tips for Testing Your Prompts

1. Start in chat: Test your prompt as a regular question before creating an automation
2. Compare with defaults: Run the same instrument with Smart Scan AI and your custom prompt to see differences
3. Iterate based on results: If the AI consistently finds no setups, your conditions may be too strict. If it finds too many, tighten your confluence requirements
4. Check the reasoning: Read the full analysis, not just the trade card. Does the AI follow your steps in order? If not, restructure your prompt
5. Track by grade: After running your automation for a week, check the A/B/C grade breakdown. If most trades are Grade C, your prompt may need more specific quality filters

Remember: The AI is only as good as your instructions. A well-structured prompt with clear steps and quality gates will consistently outperform a vague "give me trades" request.