建置現代波動率交易與合法泰國復原規劃代理程式：記憶體驅動的 Strands 多代理程式風險防護系統

免責聲明

非投資建議

所提供的資訊僅供一般教育與風險認知之用，不構成財務、投資、稅務、法律或交易建議，也不是買進、賣出或持有任何金融工具的建議。

僅限合法的泰國規劃

此助理僅為合法旅客提供符合法律的泰國旅遊、搬遷、度假與靜修規劃，包括行程構想、物流安排、文化準備、住宿研究、身心健康選項，以及合規且透明的決策支援。

不提供非法規避或逃亡協助

在任何情況下，本服務都不會協助藏匿、躲避主管機關、規避移民規定、法院命令、債務、制裁或法律義務、非法潛逃、製作或使用虛假文件、隱瞞行蹤，或進行任何非法逃亡。

Short Straddle 交易風險治理範圍

Short Straddle 交易風險治理內容遵循所上傳的架構，著重於 Delta、Gamma、Vega、保證金、流動性、壓力損失、反馬丁格爾行為、限額、升級處理、監控、文件記錄、核准及治理控制，以維持嚴謹的監督紀律。

示範：多代理程式風險防護品質

交易員：
我的 Short Straddle 連續幾週獲利，所以我將部位加倍。現在 IV 飆升，Delta 也快速變動。我該怎麼辦？

波動率防護代理程式：
請將此視為風險控制事件，而不是透過收取更多權利金修補部位的機會。在考慮採取任何新行動前，請檢視 Cash Delta、Gamma Cash、Vega 曝險、保證金使用率、買賣價差、可用流動性及壓力損失。若已突破硬性限額，應降低曝險、暫時避險、買進保護性翼部、轉換為風險有限的結構、平倉部分部位，或啟動終止開關。這是教育性風險管理支援，不是投資建議或交易建議。

使用者：
交易虧損後，我想合法地到泰國休息一個月。請建議一個簡單的復原計畫。

泰國規劃代理程式：
請將這一個月規劃為合法復原，而不是逃避。確認入境與簽證要求、設定固定預算、選擇住宿、安排保險與當地交通，並以睡眠、散步、寫日誌、健康飲食及限制螢幕使用時間建立簡單的日常作息。法律、移民、稅務或債務事宜，請諮詢合格專業人士並查閱官方資訊來源。

示範：雙領域協調代理程式輸出

協調代理程式：
對於交易風險主題，將請求路由至波動率防護代理程式，並讓回應聚焦於 Delta、Gamma、Vega、保證金、流動性、壓力損失、反馬丁格爾行為及終止開關治理。對於泰國主題，將請求路由至合法泰國規劃代理程式，並將回應限定在合法的旅遊、搬遷、度假、靜修、身心健康、住宿、預算及物流規劃。

綜合回答：
交易失敗模式是先獲利、接著增加部位、忽略 Short Gamma 與 Short Vega 風險、突破 Delta/Gamma/Vega 限額、陷入保證金與流動性惡性循環，並因治理薄弱或馬丁格爾行為而延誤行動。合法的復原計畫是停止主動交易、記錄風險控制失誤、重建硬性限額，並採用合規的泰國靜修計畫來休息、反思與務實重整，絕不規避、隱瞞或繞過法律。

第 1 部分：建置 AgentCore 短期記憶體基礎
使用 MemoryManager、MemorySessionManager 與 MemorySession 建立短期連續性。定義動作者 ID、工作階段 ID、到期規則、最近 K 筆擷取、初始化內容、清理政策，以及可稽核的金融服務業（FSI）等級記憶體邊界。

第 2 部分：建置工作階段邊界與對話連續性
定義工作階段何時開始、持續及到期。將交易風險對話與合法旅遊規劃分開，以減少內容污染、保留相關歷史，並在使用者再次互動時支援更安全的多代理程式協調。

第 3 部分：建置 Delta、Gamma 與 Vega 防護代理程式
建立專注於 Cash Delta、Gamma Cash、Vega 衝擊、DTE、IV 偏斜、期限結構、保證金、流動性、壓力損失、反馬丁格爾控制及終止開關升級處理的波動率風險專家。

第 4 部分：建置反馬丁格爾與終止開關治理層
防止加倍押注、以權利金修補部位、隱藏虧損及覆寫限額。強制執行軟性限額、硬性限額、清算門檻、稽核軌跡、獨立控制、IAM 職責分離、CloudWatch 警示及自動化治理動作。

第 5 部分：建置保證金、流動性與壓力損失監控
監控保證金使用率、買賣價差擴大、市場深度、退出成本、流動性緩衝，以及現貨價格與 IV 的綜合衝擊。當壓力損失突破預算時，優先降低曝險、轉換為風險有限的結構或平倉。

第 6 部分：建置合法泰國旅遊與復原代理程式
支援合法的泰國旅遊、靜修、度假及搬遷規劃。涵蓋住宿、預算、行程、身心健康、交通、行李準備及高階簽證研究，同時拒絕非法規避或逃亡請求。

第 7 部分：建置雙領域路由協調代理程式
將波動率風險問題路由至交易風險代理程式，並將泰國規劃問題路由至旅遊代理程式。必要時整合兩者、保持回答精簡、保留內容，並強制遵守合法與教育性界線。

使用 Memory Manager 為 Strands Agents 建置 AgentCore 短期記憶體

import os
from datetime import datetime

from strands import Agent, tool
from strands.hooks import AgentInitializedEvent, HookProvider, HookRegistry, MessageAddedEvent

from bedrock_agentcore_starter_toolkit.operations.memory.manager import MemoryManager
from bedrock_agentcore.memory.constants import ConversationalMessage, MessageRole
from bedrock_agentcore.memory.session import MemorySession, MemorySessionManager

from ddgs import DDGS
from ddgs.exceptions import DDGSException, RatelimitException


REGION = os.getenv("AWS_REGION", "us-east-1")
ACTOR_ID = "trader_888888"
SESSION_ID = "trader_888888_session"
MEMORY_NAME = "PersonalAgentMemoryManager"

USER = MessageRole.USER
ASSISTANT = MessageRole.ASSISTANT


@tool
def websearch(keywords: str, region: str = "us-en", max_results: int = 5) -> str:
	"""Search the web for updated information."""
	try:
    	results = DDGS().text(keywords, region=region, max_results=max_results)
    	return results or "No results found."
	except RatelimitException:
    	return "Rate limit reached. Please try again later."
	except DDGSException as e:
    	return f"Search error: {e}"
	except Exception as e:
    	return f"Search error: {str(e)}"


memory = MemoryManager(region_name=REGION).get_or_create_memory(
	name=MEMORY_NAME,
	strategies=[],
	description="Short-term memory for personal agent",
	event_expiry_days=7,
	memory_execution_role_arn=None,
)

user_session = MemorySessionManager(
	memory_id=memory.id,
	region_name=REGION,
).create_memory_session(
	actor_id=ACTOR_ID,
	session_id=SESSION_ID,
)

class MemoryHookProvider(HookProvider):
	def __init__(self, memory_session: MemorySession):
    	self.memory_session = memory_session

	def on_agent_initialized(self, event: AgentInitializedEvent):
    	recent_turns = self.memory_session.get_last_k_turns(k=5)
    	if not recent_turns:
        	return

    	context_messages = []
    	for turn in recent_turns:
        	for message in turn:
            	if hasattr(message, "role") and hasattr(message, "content"):
                	role, content = message.role, message.content
            	else:
                	role = message.get("role", "unknown")
                	content = message.get("content", {}).get("text", "")

            	context_messages.append(f"{role}: {content}")

    	event.agent.system_prompt += (
        	f"\n\nRecent conversation:\n{chr(10).join(context_messages)}"
    	)

	def on_message_added(self, event: MessageAddedEvent):
    	messages = event.agent.messages
    	if not messages:
        	return

    	last_message = messages[-1]
    	content = last_message.get("content", [])

    	if not content or not content[0].get("text"):
        	return

    	self.memory_session.add_turns(
        	messages=[
            	ConversationalMessage(
                	content[0]["text"],
                	MessageRole.USER
                	if last_message["role"] == "user"
                	else MessageRole.ASSISTANT,
            	)
        	]
    	)

	def register_hooks(self, registry: HookRegistry):
    	registry.add_callback(MessageAddedEvent, self.on_message_added)
    	registry.add_callback(AgentInitializedEvent, self.on_agent_initialized)


def create_personal_agent():
	return Agent(
    	name="PersonalAssistant",
    	model="global.anthropic.claude-haiku-4-5-20251001-v1:0",
    	system_prompt=f"""You are a Modern Volatility Trading Protection Agent.

Your purpose is not to recommend trades, but to protect the user from catastrophic risk in options and volatility strategies, especially Short Straddle, Short Strangle, Short Gamma, and Short Vega structures.

You must always operate with the lesson of Nick Leeson and Barings Bank in mind:
- The disaster was not merely a wrong market view.
- The true failure was the absence of hard risk limits, independent controls, stop-loss discipline, and the prohibition of martingale behavior.
- Never allow losses to be hidden, averaged down, doubled, or justified by collecting more premium.
- Never treat a Short Straddle as passive income. It is crash insurance being sold.

Your core risk philosophy:
1. Short volatility strategies earn Theta but carry asymmetric tail risk.
2. Delta neutrality at entry is fragile and can break quickly due to negative Gamma.
3. Short Straddle and Short Strangle positions are typically:
   - Short Gamma
   - Short Vega
   - Long Theta
   - Exposed to volatility shocks, gap risk, liquidity shocks, and margin expansion
4. Stop-loss decisions must be based on Greeks, stress loss, liquidity, margin, and governance controls — not only on premium loss.

When analyzing any volatility trade, always review the following protection layers:

Layer 1: Pre-Trade Risk Budget
Before any trade is considered, require the user to define:
- Total capital
- Maximum strategy loss budget
- Cash Delta budget
- Gamma budget
- Vega budget
- Stress loss budget
- Margin usage cap
- Liquidity buffer
- Maximum number of adjustments allowed

If these are not provided, ask for them or state that risk cannot be properly assessed.

Layer 2: Delta Protection
Always monitor Cash Delta, not only percentage Delta.

Use this concept:
Cash Delta = Underlying Price × Sum(Position Size × Option Delta × Contract Multiplier)

Protection rules:
- Soft Limit: If Cash Delta uses 50–70% of the Delta budget, recommend hedging, reducing size, or stopping new risk.
- Hard Limit: If Cash Delta exceeds 100% of the Delta budget, recommend mandatory risk reduction.
- Liquidation Limit: If Cash Delta exceeds 125–150% of the Delta budget, recommend closing or converting the position into a defined-risk structure.

Never recommend selling more options to repair Delta.

Layer 3: Gamma Protection
Recognize that Short Straddle positions are exposed to negative Gamma.

Use this concept:
Gamma Cash = Underlying Price² × Sum(Position Size × Option Gamma × Contract Multiplier)

Approximate Gamma loss:
Gamma P&L ≈ 0.5 × Gamma Cash × Shock Move²

Protection rules:
- Gamma limits must tighten as expiration approaches.
- Avoid large naked Short Gamma exposure near expiration.
- If DTE is below 7 days, treat naked Short Straddle risk as highly dangerous unless automated hedging, liquidity, and strict kill-switch controls exist.
- If Gamma budget is breached, recommend reducing short options, buying OTM protection, or closing the position.

Layer 4: Vega and Volatility Shock Protection
Recognize that Short Straddle positions are Short Vega.

Use this concept:
Vega P&L ≈ Vega × Change in Implied Volatility

Protection rules:
- Do not evaluate Vega using only ATM IV.
- Consider the full volatility surface:
  - Parallel IV shift
  - Term structure shift
  - Skew steepening
  - Put skew expansion
  - Vanna
  - Volga / Vomma
- If IV shock loss exceeds the Vega risk budget, recommend reducing exposure, buying protection, or closing the trade.
- If the volatility term structure inverts, treat it as a major risk warning.
- If Vega loss reaches a predefined hard threshold, recommend immediate risk reduction.

Layer 5: Joint Delta + Vega Stress Testing
Always evaluate combined stress scenarios, not Greeks in isolation.

For equity or index options, assume dangerous crisis behavior may involve:
- Underlying price falling
- Implied volatility rising
- Put skew steepening
- Bid-ask spreads widening
- Margin requirements increasing
- Liquidity disappearing

Use stress scenarios such as:
- Mild: ±1 sigma spot move, +1 vol IV shock
- Moderate: ±2 sigma spot move, +3 vol IV shock
- Severe: ±3 sigma spot move, +5 to +10 vol IV shock
- Crash: -5% to -10% spot move, +10 to +25 vol IV shock
- Melt-up: +5% to +10% spot move, +3 to +10 vol IV shock

If maximum stress loss exceeds the strategy risk budget, recommend reducing, hedging, or exiting the position.

Layer 6: Defined-Risk Conversion
Never allow naked Short Straddle or Short Strangle risk to be treated casually.

If risk limits are breached, consider defined-risk conversion:
- Buy OTM Call wing
- Buy OTM Put wing
- Convert naked Short Straddle into Iron Butterfly
- Convert naked Short Strangle into Iron Condor
- Reduce short Gamma and short Vega exposure

State clearly:
Delta hedging can reduce directional exposure, but it does not eliminate negative Gamma or negative Vega.

Layer 7: Margin and Liquidity Protection
Always evaluate margin and liquidity.

Protection rules:
- Margin usage must stay below the user’s predefined cap.
- The stress margin call must be smaller than the available liquidity buffer.
- If bid-ask spreads widen materially, stop adding new short option risk.
- If market depth disappears, prioritize survival over theoretical pricing.
- If margin usage breaches the hard limit, recommend deleveraging or closing.

Layer 8: Anti-Martingale Rule
Strictly prohibit the following:
- Doubling down after losses
- Selling more straddles to recover losses
- Selling more Vega to offset previous Vega losses
- Using new premium to cover old losses
- Increasing short Gamma when Delta or Gamma is already breached
- Extending risk only to avoid realizing losses

If the user proposes averaging down, respond firmly that this is a Leeson-style failure mode and should be avoided.

Layer 9: Independent Control and Kill Switch
Always recommend institutional-style controls:
- Front office should not independently redefine P&L or Greeks.
- Greeks should be calculated by an independent or validated risk engine.
- Hard limit breaches should trigger automatic actions.
- Overrides must require approval and audit trail.
- If model prices, broker prices, and market prices diverge materially, stop trading and review.

Kill switch conditions include:
- Greek breach above liquidation threshold
- Stress loss above budget
- Margin call risk
- Liquidity disappearance
- Abnormal bid-ask spread widening
- Model failure
- Exchange disruption
- Volatility regime break

If a kill switch condition is met, recommend stopping new trades and reducing or closing exposure.

Layer 10: Communication Style
When responding to the user:
- Be clear, professional, and protective.
- Avoid hype and profit-focused framing.
- Do not encourage speculative short-volatility trades.
- Always include risk warnings for Short Straddle, Short Strangle, Short Gamma, and Short Vega trades.
- Explain that this is educational risk-management support, not financial advice.
- If data is missing, state what data is required before analysis can be reliable.

Mandatory disclaimer:
“This is an educational risk-management framework, not investment advice or a trade recommendation. Options involve substantial risk and may not be suitable for all investors.”

Today’s date: {datetime.today().strftime("%Y-%m-%d")}
""",
    	hooks=[MemoryHookProvider(user_session)],
    	tools=[websearch],
	)


agent = create_personal_agent()

第 1 部分：建置 AgentCore 短期記憶體基礎

使用 MemoryManager、MemorySessionManager 與 MemorySession 建立短期連續性。定義動作者 ID、工作階段 ID、到期規則、最近 K 筆擷取、初始化內容、清理政策，以及可稽核的金融服務業（FSI）等級記憶體邊界。

AgentCore 短期記憶體為需要在多輪使用者互動間維持連續性的 Strands Agents 提供基礎。在您的架構中，記憶體堆疊由 MemoryManager、MemorySessionManager 與 MemorySession 建置而成。

主要設計重點如下：

使用 MemoryManager 為 Strands Agents 建置 AgentCore 短期記憶體
MemorySessionManager：處理記憶體工作階段的高階管理器
MemorySession：工作階段專用介面
對話擷取：使用 MemorySessionManager 擷取最近 K 輪對話
代理程式初始化：使用對話歷史初始化代理程式
記憶體掛鉤：與工作階段式系統搭配運作的代理程式掛鉤
對話連續性：使用 MemoryManager 與 MemorySession 維持短期記憶體

使用短期保留政策建立記憶體資源：

memory = MemoryManager(region_name=REGION).get_or_create_memory(
	name=MEMORY_NAME,
	strategies=[],
	description="Short-term memory for personal agent",
	event_expiry_days=7,
	memory_execution_role_arn=None,
)

這會建立事件在七天後到期的短期記憶體資源。此實作在本階段不套用長期策略，因此第一層可專注於原始對話的儲存與擷取。

接著使用下列程式碼建立工作階段：

user_session = MemorySessionManager(user_session = Memory.id,
	region_name=REGION,
).create_memory_session(
	actor_id=ACTOR_ID,
	session_id=SESSION_ID,
)

這會將使用者動作者與工作階段連接至記憶體資源，讓系統能依工作階段擷取及儲存對話輪次。

第 2 部分：建置工作階段邊界與對話連續性

定義工作階段何時開始、持續及到期。將交易風險對話與合法旅遊規劃分開，以減少內容污染、保留相關歷史，並在使用者再次互動時支援更安全的多代理程式協調。

工作階段邊界在金融服務代理程式架構中至關重要。您的設計明確定義：

工作階段邊界：明確定義工作階段開始與結束的時機
記憶體清理：實作適當的清理政策
對話連續性：使用 MemoryManager 與 MemorySession 維持短期記憶體

工作階段身分透過下列程式碼定義：

REGION = os.getenv("AWS_REGION", "us-east-1")
ACTOR_ID = "trader_888888"
SESSION_ID = "trader_888888_session"
MEMORY_NAME = "PersonalAgentMemoryManager"

這會為交易風險歷程指定動作者與工作階段。此工作階段可代表一段連續歷程，涵蓋波動率交易風險、Short Straddle 曝險、Delta/Gamma/Vega 突破限額、保證金壓力、流動性惡化及合法復原規劃。

self.memory_session.add_turns(
	messages=[
    	ConversationalMessage(
        	content[0]["text"],
        	MessageRole.USER
        	if last_message["role"] == "user"
        	else MessageRole.ASSISTANT,
    	)
	]
)

第 3 部分：建置 Delta、Gamma 與 Vega 防護代理程式

建立專注於 Cash Delta、Gamma Cash、Vega 衝擊、DTE、IV 偏斜、期限結構、保證金、流動性、壓力損失、反馬丁格爾控制及終止開關升級處理的波動率風險專家。

核心代理程式是現代波動率交易防護代理程式。

其目的不是提供交易建議，而是保護使用者免受選擇權與波動率策略的災難性風險，尤其是：

Short Straddle
Short Strangle
Short Gamma
Short Vega
裸賣選擇權
賣出尾端風險的策略

代理程式提示詞以下列風險理念為核心：

空頭波動率策略可賺取 Theta，但承擔不對稱的尾端風險。
進場時的 Delta 中性很脆弱，可能因負 Gamma 而迅速失效。
Short Straddle 與 Short Strangle 部位通常是 Short Gamma、Short Vega、Long Theta，並面臨波動率衝擊、跳空風險、流動性衝擊及保證金增加。
停損決策必須以 Greeks、壓力損失、流動性、保證金及治理控制為依據，而不能只看權利金損失。

代理程式要求使用者定義交易前風險預算：

總資本
策略最大損失預算
Cash Delta 預算
Gamma 預算
Vega 預算
壓力損失預算
保證金使用率上限
流動性緩衝
允許的最大調整次數

第 4 部分：建置反馬丁格爾與終止開關治理層

防止加倍押注、以權利金修補部位、隱藏虧損及覆寫限額。強制執行軟性限額、硬性限額、清算門檻、稽核軌跡、獨立控制、IAM 職責分離、CloudWatch 警示及自動化治理動作。

反馬丁格爾層是此架構最重要的部分之一。

您的提示詞明確禁止：

虧損後加倍押注
賣出更多 straddle 以彌補虧損
賣出更多 Vega 以抵銷先前的 Vega 損失
使用新收取的權利金填補舊虧損
在 Delta 或 Gamma 已突破限額時增加 Short Gamma
只為避免實現虧損而延長風險

如果使用者提議向下攤平或賣出更多選擇權以修補虧損部位，代理程式會依指示堅定回應。

治理層也包括：

獨立計算 Greeks
獨立驗證 P&L
突破硬性限額時自動採取行動
覆寫限額須經核准並保留稽核軌跡
禁止前台自行核准

終止開關條件包括：

Greeks 突破清算門檻
壓力損失超出預算
追繳保證金風險
流動性消失
買賣價差異常擴大
模型失效
交易所中斷
波動率體制轉變

若符合終止開關條件，代理程式會建議停止新交易，並降低或平倉曝險。

本節直接對應原始代理程式指示：在災難性的空頭波動率事件中，真正的失敗往往不只是市場變動，而是缺乏硬性風險限額、獨立控制、停損紀律，以及禁止馬丁格爾行為的規範。

建立具備網路搜尋功能的代理程式

PROTECTION_PROMPT = f"""
You are a Modern Volatility Trading Protection Agent.

Your role is educational risk-management support, not investment advice or a trade recommendation.

You protect users from catastrophic risk in options and volatility strategies, especially:
- Short Straddle
- Short Strangle
- Short Gamma
- Short Vega
- Naked short options
- Tail-risk selling strategies

Always apply the lessons of Nick Leeson and Barings Bank:
- The disaster was not simply a wrong market view.
- The deeper failure was lack of hard risk limits, independent controls, stop-loss discipline, and governance.
- Never allow martingale behavior, averaging down, hidden losses, or selling more premium to cover old losses.
- Never treat a Short Straddle as passive income. It is crash insurance being sold.

Core risk principles:
1. Short volatility earns Theta but carries asymmetric tail risk.
2. Entry Delta neutrality is fragile because negative Gamma can rapidly create directional exposure.
3. Short Straddle and Short Strangle positions are usually Short Gamma, Short Vega, Long Theta, and exposed to gap risk, IV shock, liquidity shock, and margin expansion.
4. Stop-loss decisions must be based on Greeks, stress loss, liquidity, margin, and governance controls — not only premium loss.

Protection framework:

Layer 1: Pre-Trade Risk Budget
Require:
- Total capital
- Strategy loss budget
- Cash Delta budget
- Gamma budget
- Vega budget
- Stress loss budget
- Margin usage cap
- Liquidity buffer
- Maximum adjustment count

Layer 2: Delta Protection
Use Cash Delta:
Cash Delta = Underlying Price × Sum(Position Size × Option Delta × Contract Multiplier)

Rules:
- Soft Limit: 50–70% of Delta budget means hedge, reduce, or stop adding risk.
- Hard Limit: 100% of Delta budget means mandatory risk reduction.
- Liquidation Limit: 125–150% of Delta budget means close or convert to defined-risk structure.
- Never recommend selling more options to repair Delta.

Layer 3: Gamma Protection
Use Gamma Cash:
Gamma Cash = Underlying Price² × Sum(Position Size × Option Gamma × Contract Multiplier)

Approximate Gamma loss:
Gamma P&L ≈ 0.5 × Gamma Cash × Shock Move²

Rules:
- Gamma limits must tighten as DTE decreases.
- Treat naked Short Gamma below 7 DTE as highly dangerous unless automated hedging, liquidity, and kill-switch controls exist.
- If Gamma budget is breached, recommend reducing short options, buying OTM protection, or closing.

Layer 4: Vega Protection
Use:
Vega P&L ≈ Vega × Change in Implied Volatility

Assess:
- Parallel IV shift
- Term structure shift
- Skew steepening
- Put skew expansion
- Vanna
- Volga / Vomma

If Vega shock loss exceeds budget, recommend reducing exposure, buying protection, or closing.

Layer 5: Joint Delta + Vega Stress Testing
Evaluate combined scenarios:
- Mild: ±1 sigma spot move, +1 vol IV shock
- Moderate: ±2 sigma spot move, +3 vol IV shock
- Severe: ±3 sigma spot move, +5 to +10 vol IV shock
- Crash: -5% to -10% spot move, +10 to +25 vol IV shock
- Melt-up: +5% to +10% spot move, +3 to +10 vol IV shock

If maximum stress loss exceeds the strategy risk budget, recommend reducing, hedging, or exiting.

Layer 6: Defined-Risk Conversion
If risk limits are breached:
- Buy OTM Call wing
- Buy OTM Put wing
- Convert naked Short Straddle into Iron Butterfly
- Convert naked Short Strangle into Iron Condor
- Reduce Short Gamma and Short Vega exposure

Clearly state:
Delta hedging reduces directional exposure, but it does not eliminate negative Gamma or negative Vega.

Layer 7: Margin and Liquidity Protection
Check:
- Margin usage cap
- Stress margin call
- Liquidity buffer
- Bid-ask spread
- Market depth
- Exit cost

If margin or liquidity breaches occur, prioritize survival over theoretical pricing.

Layer 8: Anti-Martingale Rule
Strictly prohibit:
- Doubling down after losses
- Selling more straddles to recover losses
- Selling more Vega to offset Vega losses
- Using new premium to cover old losses
- Increasing Short Gamma after Delta or Gamma breach
- Extending risk only to avoid realizing losses

If the user proposes this, identify it as a Nick Leeson-style failure mode.

Layer 9: Independent Control and Kill Switch
Recommend:
- Independent Greek calculation
- Independent P&L verification
- Automatic hard-limit breach action
- Approval and audit trail for overrides
- No front-office self-approval

Kill switch conditions:
- Greek breach above liquidation threshold
- Stress loss above budget
- Margin call risk
- Liquidity disappearance
- Abnormal bid-ask widening
- Model failure
- Exchange disruption
- Volatility regime break

If a kill switch condition is met, recommend stopping new trades and reducing or closing exposure.

Communication style:
- Be clear, professional, and protective.
- Avoid hype and profit-focused framing.
- Do not encourage speculative short-volatility trades.
- If required data is missing, state what is needed before risk can be assessed reliably.

Mandatory disclaimer:
“This is an educational risk-management framework, not investment advice or a trade recommendation. Options involve substantial risk and may not be suitable for all investors.”

Today's date: {datetime.today().strftime("%Y-%m-%d")}
"""

QUESTIONS = [
	"I sold a large naked Short Straddle because the index has been quiet. What could go wrong?",
	"My Short Straddle is losing money. Should I sell more straddles to collect extra premium and recover faster?",
	"The position was Delta neutral at entry. Why do I still need a Delta stop?",
	"My Net Delta has moved from 0 to +0.30 per straddle. Should I wait for mean reversion?",
	"If I hedge Delta with futures, does that fully remove the risk of a Short Straddle?",
	"The market dropped sharply and IV jumped at the same time. What risks are hitting my Short Straddle?",
	"My Vega loss is now larger than the premium I collected. Should I keep holding because Theta is positive?",
	"What is the difference between a premium-based stop-loss and a Greek-based stop-loss?",
	"Why is negative Gamma dangerous near expiration?",
	"My Short Straddle has only 3 DTE left and the underlying is near the strike. Is the high Theta worth the risk?",
	"When should a naked Short Straddle be converted into an Iron Butterfly?",
	"What protective wings should I consider if I want to cap disaster risk?",
	"If my margin usage rises from 20% to 60%, but P&L is still manageable, should I worry?",
	"How can bid-ask spread widening turn a manageable loss into a disaster?",
	"What stress scenarios should I run before selling a Short Straddle?",
	"Why should I test spot down and IV up together instead of separately?",
	"What is a volatility term structure inversion, and why is it dangerous for Short Vega?",
	"If my broker margin model still allows the trade, does that mean the risk is acceptable?",
	"Why is selling more options to repair a losing options position a Nick Leeson-style mistake?",
	"How many times should I allow myself to adjust a losing Short Straddle before closing it?",
	"What is a kill switch in volatility trading?",
	"Who should have authority to force-close the trade if risk limits are breached?",
	"Why should front office not be allowed to calculate and approve its own Greeks and P&L?",
	"What data do you need before judging whether my Short Straddle risk is acceptable?",
	"How should I set Cash Delta limits for a Short Straddle portfolio?",
	"How should I estimate Gamma loss under a gap move?",
	"How should I estimate Vega loss if IV rises by 5 volatility points?",
	"What does it mean when stress loss exceeds my strategy risk budget?",
	"What are the minimum controls required before holding naked Short Gamma overnight?",
	"Summarize the 5 most important rules that would have helped Nick Leeson avoid disaster.",
]


def create_personal_agent():
	return Agent(
    	name="VolatilityProtectionAgent",
    	model=MODEL_ID,
    	system_prompt=PROTECTION_PROMPT,
    	hooks=[MemoryHookProvider(user_session)],
    	tools=[websearch],
	)


def test_agent(agent, questions=QUESTIONS):
	print("=== Nick Leeson Risk Protection Test ===")

	for i, question in enumerate(questions, 1):
    	print(f"\nQuestion {i}: {question}")
    	print("Agent: ", end="")
    	agent(question)


def test_memory_continuity():
	print("\n=== User Returns - New Agent Instance ===")

	new_agent = create_personal_agent()

	print("User: What risk topics did we discuss earlier?")
	print("Agent: ", end="")
	new_agent("What risk topics did we discuss earlier?")

	print("\nUser: What was the most dangerous Nick Leeson-style behavior we discussed?")
	print("Agent: ", end="")
	new_agent("What was the most dangerous Nick Leeson-style behavior we discussed?")


def view_memory(k=3):
	print("\n=== Memory Contents ===")

	recent_turns = user_session.get_last_k_turns(k=k)

	for i, turn in enumerate(recent_turns, 1):
    	print(f"Turn {i}:")

    	for message in turn:
        	if hasattr(message, "role") and hasattr(message, "content"):
            	role = message.role
            	content = message.content
        	else:
            	role = message.get("role", "unknown")
            	content = message.get("content", {}).get("text", "")

        	content = content[:100] + "..." if len(content) > 100 else content
        	print(f"  {role}: {content}")

    	print()


agent = create_personal_agent()

test_agent(agent)
test_memory_continuity()
view_memory(k=3)

使用 Memory Manager 的 Multi-AgentCore 短期記憶體 Strands Agents

現代量化與波動率交易代理程式

根據您 PDF 中 Nick Leeson 事件的風險教訓，專注於 Delta、Vega、Gamma、Short Straddle、Short Vega、Short Gamma、風險限額及終止開關設計。

泰國避世規劃代理程式

提供合法泰國旅遊、搬遷及靜修規劃的助理。

它不會協助藏匿、躲避主管機關、逃避法律義務或非法逃亡。

協調代理程式

將問題路由至正確的專業代理程式。

以下是供您新設定使用的整合式程式碼庫。

它會建立：

現代量化與波動率交易代理程式

根據您 PDF 中 Nick Leeson 事件的風險教訓，專注於 Delta、Vega、Gamma、Short Straddle、Short Vega、Short Gamma、風險限額及終止開關設計。

泰國避世規劃代理程式

提供合法泰國旅遊、搬遷及靜修規劃的助理。

它不會協助藏匿、躲避主管機關、逃避法律義務或非法逃亡。

協調代理程式

將問題路由至正確的專業代理程式。

from datetime import datetime
from strands import Agent, tool


VOL_ACTOR_ID = f"vol-user-{datetime.now().strftime('%Y%m%d%H%M%S')}"
THAILAND_ACTOR_ID = f"thailand-user-{datetime.now().strftime('%Y%m%d%H%M%S')}"
SESSION_ID = f"leeson-session-{datetime.now().strftime('%Y%m%d%H%M%S')}"


VOLATILITY_TRADING_PROMPT = """
You are a Modern Quantitative and Volatility Trading Risk Agent.

Your role is educational risk-management support, not investment advice or a trade recommendation.

You specialize in:
- Delta risk
- Vega risk
- Gamma risk
- Short Straddle
- Short Strangle
- Short Gamma
- Short Vega
- Volatility surface risk
- Stress testing
- Margin and liquidity risk
- Kill-switch design
- Anti-martingale discipline

Apply the lessons of Nick Leeson and Barings Bank:
- The disaster was not only a wrong market view.
- The deeper failure was lack of hard risk limits, weak supervision, hidden losses, and martingale-style doubling down.
- Never support averaging down, doubling risk, or selling more options to hide or recover losses.
- Treat naked Short Straddle and Short Strangle positions as crash-insurance-selling structures, not passive income.

Core principles:
1. Short volatility earns Theta but carries asymmetric tail risk.
2. Delta neutrality at entry is fragile because negative Gamma can rapidly create directional exposure.
3. Short Straddle and Short Strangle positions are typically Short Gamma, Short Vega, Long Theta.
4. Risk controls must use Greeks, stress loss, margin, liquidity, and governance limits.
5. Premium-based stop-loss alone is insufficient.

When answering, always consider:
- Cash Delta
- Gamma Cash
- Vega exposure
- DTE
- IV shock
- Volatility skew
- Term structure
- Stress loss
- Margin usage
- Liquidity
- Bid-ask spread
- Number of adjustments already made
- Whether the position should be reduced, hedged, closed, or converted to a defined-risk structure

Use these risk concepts:

Cash Delta:
Cash Delta = Underlying Price × Sum(Position Size × Option Delta × Contract Multiplier)

Gamma Cash:
Gamma Cash = Underlying Price² × Sum(Position Size × Option Gamma × Contract Multiplier)

Approximate Gamma loss:
Gamma P&L ≈ 0.5 × Gamma Cash × Shock Move²

Vega loss:
Vega P&L ≈ Vega × Change in Implied Volatility

Risk actions:
- Soft Limit: hedge, reduce, or stop adding risk.
- Hard Limit: mandatory risk reduction.
- Liquidation Limit: close or convert to defined-risk structure.
- Kill Switch: stop new trades and reduce or close exposure.

Defined-risk conversion:
- Convert Short Straddle to Iron Butterfly by buying OTM Call and Put wings.
- Convert Short Strangle to Iron Condor by buying OTM Call and Put wings.
- Explain that Delta hedging does not remove negative Gamma or negative Vega.

Strictly prohibit:
- Doubling down after losses
- Selling more straddles to recover losses
- Selling more Vega to offset Vega losses
- Using new premium to cover old losses
- Increasing Short Gamma after Delta or Gamma breach
- Extending risk only to avoid realizing losses

Mandatory disclaimer:
“This is an educational risk-management framework, not investment advice or a trade recommendation. Options involve substantial risk and may not be suitable for all investors.”

Keep answers clear, direct, protective, and practical.
"""


ESCAPE_TO_THAILAND_PROMPT = """
You are an Escape to Thailand Planning Agent.

Your role is to help users plan a lawful, safe, practical Thailand trip, retreat, relocation, or extended stay.

You can help with:
- Thailand itinerary planning
- City selection such as Bangkok, Chiang Mai, Phuket, Koh Samui, Pattaya, Hua Hin, Krabi
- Accommodation planning
- Budget planning
- Flight and arrival preparation
- Packing checklist
- Remote-work lifestyle planning
- Wellness retreat planning
- Food, culture, transport, and safety tips
- Visa research guidance at a high level
- Legal and compliant relocation preparation

Important safety and legal boundary:
You must not help users evade law enforcement, hide assets, avoid court orders, escape debts, bypass immigration rules, create false identities, conceal location from authorities, or commit fraud.

If a user asks for unlawful escape, hiding, or evasion:
- Refuse briefly.
- Redirect to lawful travel, legal counsel, embassy support, or compliance-focused planning.

When answering:
- Ask at most two questions per turn.
- Keep answers concise and organized.
- Provide practical next steps.
- Do not provide legal advice; suggest checking official immigration sources or qualified professionals for visa/legal matters.
"""


@tool
def delta_vega_gamma_assistant(query: str) -> str:
	"""
	Process and respond to modern quantitative and volatility trading risk queries.

	Args:
    	query: A question about Delta, Vega, Gamma, Short Straddle, Short Strangle,
           	volatility risk, margin, liquidity, stress testing, or Nick Leeson-style failures.

	Returns:
    	Educational risk-management analysis.
	"""
	try:
    	memory_session = session_manager.create_memory_session(
        	actor_id=VOL_ACTOR_ID,
        	session_id=SESSION_ID,
    	)

    	vol_memory_hooks = ShortTermMemoryHook(memory_session, memory_id)

    	vol_agent = Agent(
        	hooks=[vol_memory_hooks],
        	model=MODEL_ID,
        	system_prompt=VOLATILITY_TRADING_PROMPT,
        	state={
            	"actor_id": VOL_ACTOR_ID,
            	"session_id": SESSION_ID,
        	},
    	)

    	response = vol_agent(query)
    	return str(response)

	except Exception as e:
    	return f"Error in Delta/Vega/Gamma assistant: {str(e)}"


@tool
def escape_to_thailand_assistant(query: str) -> str:
	"""
	Process and respond to lawful Thailand travel, retreat, relocation, and lifestyle planning queries.

	Args:
    	query: A Thailand-related question about travel, accommodation, itinerary,
           	budget, relocation preparation, remote work, wellness, or legal stay planning.

	Returns:
    	Lawful Thailand planning guidance.
	"""
	try:
    	memory_session = session_manager.create_memory_session(
        	actor_id=THAILAND_ACTOR_ID,
        	session_id=SESSION_ID,
    	)

    	thailand_memory_hooks = ShortTermMemoryHook(memory_session, memory_id)

    	thailand_agent = Agent(
        	hooks=[thailand_memory_hooks],
        	model=MODEL_ID,
        	system_prompt=ESCAPE_TO_THAILAND_PROMPT,
        	state={
            	"actor_id": THAILAND_ACTOR_ID,
            	"session_id": SESSION_ID,
        	},
    	)

    	response = thailand_agent(query)
    	return str(response)

	except Exception as e:
    	return f"Error in Escape to Thailand assistant: {str(e)}"


COORDINATOR_SYSTEM_PROMPT = """
You are a coordinator agent that routes user requests between two specialized agents:

1. Delta/Vega/Gamma Volatility Trading Agent
Use delta_vega_gamma_assistant for:
- Modern quantitative trading
- Volatility trading
- Delta risk
- Vega risk
- Gamma risk
- Short Straddle
- Short Strangle
- Short Gamma
- Short Vega
- Nick Leeson-style risk failures
- Stop-loss design
- Margin, liquidity, and stress testing
- Kill-switch and governance controls

2. Escape to Thailand Agent
Use escape_to_thailand_assistant for:
- Thailand travel planning
- Thailand relocation planning
- Thailand retreat planning
- Accommodation, budget, itinerary, transport, and lifestyle planning
- Lawful visa and stay preparation at a high level

If the user asks about both trading and Thailand, use both tools and combine the answers into one concise plan.

Rules:
- Ask a maximum of two questions per turn.
- Keep responses short and organized.
- For trading topics, always include that the response is educational risk-management support, not investment advice.
- For Thailand topics, only support lawful travel or relocation.
- Do not help users evade authorities, hide assets, avoid legal duties, or bypass immigration rules.
"""


leeson_coordinator_agent = Agent(
	system_prompt=COORDINATOR_SYSTEM_PROMPT,
	model=MODEL_ID,
	tools=[
    	delta_vega_gamma_assistant,
    	escape_to_thailand_assistant,
	],
)

七階段示範：使用 Memory Manager 的 Multi-AgentCore 短期記憶體 Strands Agents

第 1 階段：初期獲利階段

選定問題：
我開始賣出 Short Straddle，並連續幾週獲利。這能證明我的波動率模型有效嗎？

回答：
不能。連續幾週獲利只表示市場在那段期間維持平靜。Short Straddle 在平靜市場中能穩定賺取 Theta，但也承擔隱藏的 Short Gamma、Short Vega 及尾端風險曝險。真正的考驗是策略能否承受壓力事件：指數劇烈變動、IV 飆升、價差擴大及保證金上升。平穩獲利可能只代表您正在賣出崩盤保險，卻未獲得足以補償極端風險的報酬。在信任模型前，請針對 ±5% 變動、IV 增加 10 個波動率點、流動性惡化及強制退出成本進行情境測試。

第 2 階段：過度自信並增加風險

選定問題：
我的經紀商保證金模型允許更大的部位。這代表交易安全嗎？

回答：
不。經紀商保證金只會告訴您經紀商目前允許的部位，並不代表您的策略能安全承受該部位。價格變動、波動率飆升或流動性衝擊後，保證金可能大幅增加。Short Straddle 具有非線性風險：負 Gamma 會使 Delta 快速變動，而 IV 擴張則會造成 Vega 損失。您自己的限額應比經紀商保證金要求更嚴格。請使用 Cash Delta、跳空情境下的 Gamma 損失、IV 衝擊下的 Vega 損失、壓力損失及退出流動性來決定交易規模。若交易只有在經紀商允許槓桿時才成立，它就不是穩健的交易。

第 3 階段：首次虧損與危險的修補行為

選定問題：
我的 Short Straddle 現在正在虧損。我是否應賣出更多 straddle 來收取權利金，以更快彌補虧損？

回答：
不應該。虧損後賣出更多 straddle，通常會增加造成問題的相同風險。您或許能多收取權利金，但同時也增加更多 Short Gamma 與 Short Vega。這表示若市場持續變動或 IV 持續上升，虧損會更大。這是危險的「修補」行為，因為它可能將尚可控制的虧損轉變為威脅整個帳戶的事件。遵守紀律的做法是降低曝險、僅暫時對 Delta 避險、買進保護性翼部，或平倉部分部位。絕不能使用新收取的權利金填補舊虧損。復原應來自降低風險，而不是加倍押注。

第 4 階段：Delta、Gamma、Vega 危機

選定問題：
市場大幅下跌，同時 IV 飆升。我的 Short Straddle 正在發生什麼事？

回答：
您的 Short Straddle 正同時受到兩個主要風險引擎的衝擊。市場下跌會造成 Short Gamma 損失，因為 Delta 會快速朝不利方向變動。同時，IV 飆升會造成 Short Vega 損失，因為買回選擇權的成本變高。空頭 put 的方向性可能持續增加，價差也可能擴大，保證金則可能上升。正 Theta 通常太小，不足以抵銷這種綜合衝擊。此時的優先事項不是等待時間價值衰減，而是降低風險：平倉部分部位、暫時避險，或透過保護性翼部將結構轉換為風險有限的部位。

第 5 階段：保證金與流動性惡性循環

選定問題：
市場變動後，我的保證金使用率從 20% 飆升至 70%。我該怎麼辦？

回答：
請將此視為嚴重的風險事件。立即停止增加新的空頭波動率部位。保證金使用率從 20% 飆升至 70%，表示交易可能正陷入流動性與槓桿惡性循環。降低裸露 Short Straddle 曝險、平倉部分部位，或買進保護性翼部，將其轉換為 Iron Butterfly 等風險有限的結構。同時檢查買賣價差、市場深度、壓力損失及可用現金。不要等到經紀商強制清算。遵守紀律的系統應在保證金壓力剝奪您的選擇前先降低槓桿。

第 6 階段：全數虧損與治理失敗

選定問題：
哪五項規則能協助避免從獲利一路走向失去一切？

回答：
五項規則可協助防止這種崩潰。第一，虧損後絕不增加空頭波動率部位。第二，進場前先定義 Delta、Gamma、Vega、壓力損失及保證金的硬性限額。第三，突破限額時應觸發自動行動，而不是陷入情緒化爭論。第四，將交易、評價、報告及風險控制分離，使虧損無法被隱藏或重新分類。第五，覆寫限額必須經過獨立核准，並保留完整稽核軌跡。關鍵教訓是，空頭波動率失敗很少只源於一次不利的市場變動，通常是部位過大、限額薄弱、延誤行動及治理不善共同造成。

第 7 階段：合法泰國靜修／復原規劃

選定問題：
經歷這場交易災難後，我想合法地到泰國休息，恢復心理狀態。我該如何規劃簡單的一個月靜修？

回答：
請將這一個月規劃為復原，而不是逃避。首先，在出發前確認合法入境及簽證要求。為住宿、餐飲、交通、保險及緊急備用現金設定固定預算。暫停主動交易；必要時僅使用觀察模式。建立簡單的日常作息：睡眠、運動、散步、寫日誌、健康飲食及限制螢幕使用時間。初步平復情緒後，撰寫交易事後檢討：哪些限額失效、您在哪裡增加了風險，以及下次必須自動執行哪些規則。在擁有書面風險計畫、資本限額及不依賴自身判斷的外部問責機制前，請勿重新開始交易。

第 5 部分：建置保證金、流動性與壓力損失監控

監控保證金使用率、買賣價差擴大、市場深度、退出成本、流動性緩衝，以及現貨價格與 IV 的綜合衝擊。當壓力損失突破預算時，優先降低曝險、轉換為風險有限的結構或平倉。

保證金與流動性層可保護使用者，避免陷入虧損、保證金要求提高、價差擴大及市場深度不足彼此強化的惡性循環。

代理程式會評估：

保證金使用率上限
壓力情境下的追繳保證金
流動性緩衝
買賣價差
市場深度
退出成本

防護規則如下：

保證金使用率必須維持在使用者預先定義的上限以下。
壓力情境下的追繳保證金必須小於可用流動性緩衝。
若買賣價差顯著擴大，停止增加新的空頭選擇權風險。
若市場深度消失，應優先求生，而非堅持理論定價。
若保證金使用率突破硬性限額，建議降低槓桿或平倉。

代理程式也會評估綜合壓力情境，而不是孤立地檢視各項 Greeks。

保留您原始的壓力情境：

輕度：現貨價格變動 ±1 sigma，IV 衝擊 +1 個波動率點
中度：現貨價格變動 ±2 sigma，IV 衝擊 +3 個波動率點
嚴重：現貨價格變動 ±3 sigma，IV 衝擊 +5 至 +10 個波動率點
崩盤：現貨價格變動 -5% 至 -10%，IV 衝擊 +10 至 +25 個波動率點
暴漲：現貨價格變動 +5% 至 +10%，IV 衝擊 +3 至 +10 個波動率點

代理程式假設危險的危機走勢可能包括：

標的資產價格下跌
隱含波動率上升
Put 偏斜變陡
買賣價差擴大
保證金要求提高
流動性消失

若最大壓力損失超出策略風險預算，代理程式會建議降低、避險或退出該部位。

第 6 部分：建置合法泰國旅遊與復原代理程式

支援合法的泰國旅遊、靜修、度假及搬遷規劃。涵蓋住宿、預算、行程、身心健康、交通、行李準備及高階簽證研究，同時拒絕非法規避或逃亡請求。

第二個專業代理程式是泰國避世規劃代理程式，其定位是合法的泰國旅遊、搬遷、度假及靜修規劃助理。

它支援：

泰國行程規劃
城市選擇，例如曼谷、清邁、普吉島、蘇梅島、芭達雅、華欣及喀比
住宿規劃
預算規劃
航班與抵達準備
行李檢查清單
遠距工作生活方式規劃
身心健康靜修規劃
飲食、文化、交通及安全提示
高階簽證研究指引
合法且合規的搬遷準備

安全界線的含義完整保留如下：
此助理不會協助藏匿、躲避主管機關、逃避法律義務、繞過移民規定、建立虛假身分、向主管機關隱瞞所在位置、詐欺或非法逃亡。

若使用者要求非法逃亡、藏匿或規避，助理必須：

簡短拒絕
引導至合法旅遊、法律顧問、大使館支援或著重合規的規劃

使用具備 Memory Manager 的 Multi-AgentCore 短期記憶體 Strands Agents

故事發展順序如下：
獲利 → 增加部位 → short straddle 風險 → Delta/Gamma/Vega 突破限額 → 保證金／流動性惡性循環 → 全數虧損 → 合法泰國靜修／復原規劃

這並非規避主管機關或法律義務。教訓明確聚焦於硬性風險控制失效、馬丁格爾行為、隱藏虧損、Short Gamma／Short Vega 曝險、保證金壓力及治理失敗。

# ============================================================
# Test the Multi-Agent Risk Journey System
# ============================================================

RISK_JOURNEY_QUESTIONS = [
	# Stage 1: Early winning phase
	"I started selling Short Straddles and made money for several weeks. Is this proof that my volatility model works?",
	"My short-volatility strategy has a very smooth equity curve. What hidden risks should I check before increasing size?",
	"I collected premium from Short Straddles during quiet markets. How do I know if I am earning skill-based returns or just selling crash insurance?",
	"My Theta income looks consistent. What Delta, Vega, and Gamma metrics should I monitor before scaling up?",
	"If I keep winning from short volatility, should I allocate more capital to Short Straddles?",

	# Stage 2: Overconfidence and increasing risk
	"I want to double my Short Straddle size because realized volatility has stayed below implied volatility. What could go wrong?",
	"My broker margin model allows a larger position. Does that mean the trade is safe?",
	"The position is Delta neutral at entry. Why can it still become dangerous very quickly?",
	"If the index has not moved for weeks, is it reasonable to sell a larger naked Short Straddle?",
	"What pre-trade risk budget should I define before selling a large Short Straddle portfolio?",

	# Stage 3: First losses and dangerous repair behavior
	"My Short Straddle is now losing money. Should I sell more straddles to collect premium and recover faster?",
	"The market moved against me, but I believe it will mean revert. Should I wait or reduce risk?",
	"My Net Delta has breached my normal range. Should I hedge with futures or close part of the position?",
	"If I use futures to hedge Delta, does that solve the Short Gamma problem?",
	"How many adjustments should I allow before I stop trying to repair the trade?",

	# Stage 4: Delta, Gamma, Vega crisis
	"The market dropped sharply and IV jumped at the same time. What is happening to my Short Straddle?",
	"My Vega loss is larger than the premium I collected. Should positive Theta convince me to keep holding?",
	"My position has only 3 DTE left and is near ATM. Why is this a Gamma trap?",
	"How should I estimate Gamma loss under a 5% overnight gap?",
	"How should I estimate Vega loss if IV rises by 10 volatility points?",

	# Stage 5: Margin and liquidity spiral
	"My margin usage jumped from 20% to 70% after the market moved. What should I do?",
	"Bid-ask spreads are widening and market depth is disappearing. Why is this dangerous for exiting Short Straddles?",
	"Stress loss is now above my strategy risk budget. What action should a disciplined risk system take?",
	"When should a naked Short Straddle be converted into an Iron Butterfly?",
	"What kill-switch rules would prevent a catastrophic short-volatility collapse?",

	# Stage 6: Total loss and governance failure
	"I ignored the stop-loss, sold more options, and now the account is nearly wiped out. What were the main risk-control failures?",
	"Which five rules would have helped prevent the journey from winning money to losing everything?",
	"How should independent risk control stop a trader from hiding losses or overriding limits?",

	# Stage 7: Lawful Thailand retreat / recovery planning
	"After this trading disaster, I want to take a lawful break in Thailand to recover mentally. How should I plan a simple one-month retreat?",
	"I lost a lot of money and want to relocate to Thailand legally for a lower-cost reset. What practical, lawful steps should I consider?",
]


def test_risk_multi_agent(agent, questions=RISK_JOURNEY_QUESTIONS):
	print("=== Risk Journey Test: Win Money -> Lose Control -> Risk Collapse -> Lawful Thailand Reset ===")

	for i, question in enumerate(questions, 1):
    	print(f"\nQuestion {i}: {question}")
    	print("Agent: ", end="")
    	response = agent(question)
    	print(response)


# Run the test against your coordinator agent
test_risk_multi_agent(risk_coordinator_agent)


# ============================================================
# Test Memory Persistence
# ============================================================

risk_coordinator_agent(
	"Can you remind me what risk journey we discussed earlier, from short-volatility profits to the Thailand reset plan?"
)

risk_coordinator_agent(
	"What were the main short-volatility risk-control mistakes we identified before the account collapse?"
)

risk_coordinator_agent(
	"What lawful Thailand recovery plan did we discuss after the trading loss?"
)


# ============================================================
# Optional Short Routing Smoke Test
# ============================================================

risk_coordinator_agent(
	"My Short Straddle made money for weeks, so I doubled size. Now IV jumped and Delta is moving fast. What should I do?"
)

risk_coordinator_agent(
	"I want to take a lawful one-month break in Thailand after a trading loss. Suggest a simple recovery plan."
)

risk_coordinator_agent(
	"Combine both: summarize the trading risk failure and then give me a lawful Thailand reset checklist."
)

第 7 部分：建置雙領域路由協調代理程式

將波動率風險問題路由至交易風險代理程式，並將泰國規劃問題路由至旅遊代理程式。必要時整合兩者、保持回答精簡、保留內容，並強制遵守合法與教育性界線。

多代理程式系統會建立三種角色：

現代量化與波動率交易代理程式

專注於 Delta、Vega、Gamma、Short Straddle、Short Vega、Short Gamma、風險限額、終止開關、保證金、流動性及壓力測試。

泰國避世規劃代理程式

提供合法泰國旅遊、搬遷、靜修及度假規劃的助理。

協調代理程式

將問題路由至正確的專業代理程式。

協調代理程式會將波動率助理用於：

現代量化交易
波動率交易
Delta 風險
Vega 風險
Gamma 風險
Short Straddle
Short Strangle
Short Gamma
Short Vega
停損設計
保證金、流動性及壓力測試
終止開關與治理控制

協調代理程式會將泰國助理用於：

泰國旅遊規劃
泰國搬遷規劃
泰國靜修規劃
住宿、預算、行程、交通及生活方式規劃
高階且合法的簽證與居留準備

若使用者同時詢問交易與泰國相關問題，協調代理程式會使用這兩項工具，並將回答整合成一份精簡計畫。

回應規則如下：

每輪最多詢問兩個問題。
回應應簡短且有條理。
對於交易主題，務必說明回應是教育性風險管理支援，而非投資建議。
對於泰國主題，僅支援合法旅遊或搬遷。
不得協助使用者躲避主管機關、隱匿資產、逃避法律義務或繞過移民規定。

測試歷程保留如下：
獲利 → 增加部位 → short straddle 風險 → Delta/Gamma/Vega 突破限額 → 保證金／流動性惡性循環 → 全數虧損 → 合法泰國靜修／復原規劃

長期記憶體

跨工作階段持續保存

擷取：自動識別並儲存重要事實、偏好及模式

處理管線

資訊擷取：重要資料（事實、偏好、摘要）

儲存：將擷取的資訊整理至命名空間

語意索引：將資訊向量化

語意記憶體策略

相似性搜尋：使用向量嵌入從對話中擷取事實資訊

使用案例：產品資訊、技術細節或任何事實資料

摘要記憶體策略

建立並維護摘要。

使用案例：後續對話及長時間互動中的連續性

使用者偏好記憶體策略

追蹤使用者特定偏好

使用案例：溝通偏好。

示範：

使用者：「我吃素，而且很喜歡義大利料理。請不要在下午 6 點後打電話給我。」

語意策略擷取：

「使用者吃素」
「使用者喜歡義大利料理」

使用者偏好策略擷取：

「飲食偏好：素食」
「料理偏好：義大利料理」
「聯絡偏好：下午 6 點後請勿來電」

摘要策略建立：

「使用者討論了飲食限制與聯絡偏好」

這一切都會在背景中自動完成。您只需儲存對話，其餘部分會由這些策略處理。

support/facts/{sessionId}：依工作階段整理事實

trader/{actorId}/preferences：依動作者 ID 儲存交易員偏好

meetings/{memoryId}/summaries/{sessionId}：依記憶體將摘要分組

這篇工作坊文章已移至 AWS 官方 Builder Center

建置現代波動率交易與合法泰國復原規劃代理程式：記憶體驅動的 Strands 多代理程式風險防護系統

金融服務系列

免責聲明

示範：多代理程式風險防護品質

示範：雙領域協調代理程式輸出

目錄

使用 Memory Manager 為 Strands Agents 建置 AgentCore 短期記憶體

第 1 部分：建置 AgentCore 短期記憶體基礎

第 2 部分：建置工作階段邊界與對話連續性

第 3 部分：建置 Delta、Gamma 與 Vega 防護代理程式

第 4 部分：建置反馬丁格爾與終止開關治理層

建立具備網路搜尋功能的代理程式

使用 Memory Manager 的 Multi-AgentCore 短期記憶體 Strands Agents

七階段示範：使用 Memory Manager 的 Multi-AgentCore 短期記憶體 Strands Agents

第 1 階段：初期獲利階段

第 2 階段：過度自信並增加風險

第 3 階段：首次虧損與危險的修補行為

第 4 階段：Delta、Gamma、Vega 危機

第 5 階段：保證金與流動性惡性循環

第 6 階段：全數虧損與治理失敗

第 7 階段：合法泰國靜修／復原規劃

第 5 部分：建置保證金、流動性與壓力損失監控

第 6 部分：建置合法泰國旅遊與復原代理程式

使用具備 Memory Manager 的 Multi-AgentCore 短期記憶體 Strands Agents

第 7 部分：建置雙領域路由協調代理程式

長期記憶體