starchild-ai-agent/@3182/regime-aware-entry-engine

Regime-Aware Entry Engine

When to use

Use this workflow when the user asks:

• "Is this dip a buy?"
• "What are the opportunities after a selloff?"
• "Give me entry/add/stop/target levels"
• "Build me a tactical watchlist"

Works for equities/ETFs (US/HK/A-share) with market data support.

Output format (mandatory)

Always produce a table with:

• Ticker
• First Buy
• Add1 (-8%)
• Add2 (-15%)
• Stop
• Target
• Upside
• Regime Tag (RISK_ON, RISK_OFF, PANIC)
• Action (BUY_NOW, WATCH, AVOID)

Workflow

1) Detect regime first

Use broad-market proxies before single-name calls:

• Growth beta proxy (e.g., QQQ)
• Broad market proxy (e.g., SPY)
• Sector stress proxy (e.g., SOXX/SMH when tech-led)
• Defensive relative strength (e.g., XLU/XLV vs XLK)

Classify:

• RISK_ON: breadth stable, growth leading, no stress spike
• RISK_OFF: broad pullback, cyclicals weak, defensives holding up
• PANIC: disorderly selloff / correlation spike / large intraday dislocations

2) Build candidate pool

Start with user’s watchlist first. If not provided, build from:

• prior winners with healthy pullback
• sector leaders with intact structure
• avoid low-liquidity names in stress regimes

3) Score each candidate

Use a simple composite score (0–100):

• Trend quality (35%): still above key medium trend context; no structural break
• Pullback quality (30%): meaningful discount from recent high, but not freefall
• Risk efficiency (20%): stop distance vs expected upside
• Liquidity/Execution (15%): spreads/volume/volatility practicality

4) Generate execution levels

• First Buy: current value zone (no chasing)
• Add1: approx -8% from first buy reference
• Add2: approx -15% from first buy reference
• Stop: invalidation level (structure break)
• Target: base-case recovery objective
• Upside: (Target / current - 1)

5) Apply regime guardrails

• In RISK_ON: normal position sizing, allow 2-stage adds
• In RISK_OFF: reduce first tranche size, prioritize quality balance sheets
• In PANIC: smaller probes only; keep larger cash buffer; avoid forced averaging

6) Produce clear actions

• BUY_NOW: in zone + structure acceptable
• WATCH: close but not yet in range / waiting confirmation
• AVOID: structure broken, governance risk, or asymmetry poor

Messaging style

• Keep it execution-first, concise, and numeric.
• Do not promise certainty.
• Always include a one-line risk note: "Not investment advice; follow stop discipline."

Quality checks before finalizing

• Levels are internally consistent (Add2 < Add1 < First Buy)
• Stop is below buy ladders for long setups
• Upside is computed from current price, not stale references
• Action tag matches regime + structure

Optional extension (risk layer)

If governance/event risk is detected (management/legal/disclosure shock), force downgrade:

• Action cannot be BUY_NOW until risk is clarified
• Cap size at half normal risk budget

Reuse pattern

For repeated daily usage:

1. Pull latest broad market + watchlist quotes
2. Recompute regime
3. Update table only where trigger state changed
4. Push only changed rows to avoid alert spam