"""eCommerce supervisor agent (sample) for Amazon Bedrock AgentCore Runtime.

Versions / setup:
  - Python 3.12+
  - strands-agents >= 1.x
  - bedrock-agentcore Runtime HTTP contract (BedrockAgentCoreApp)
  - Model id below is a pin for docs; swap for your region/account

This file routes intents to specialist Runtime ARNs (or local stubs).
It does NOT call refund/cancel/inventory write tools directly — those stay
behind Gateway Policy on the specialists.

Demo only: fake routing; replace invoke_specialist with real Runtime invoke.
"""

from __future__ import annotations

import json
import os
from typing import Any

from bedrock_agentcore.runtime import BedrockAgentCoreApp
from strands import Agent, tool
from strands.models.bedrock import BedrockModel

MODEL_ID = os.environ.get(
    "BEDROCK_MODEL_ID",
    "global.anthropic.claude-sonnet-4-5-20250929-v1:0",
)

# Map logical specialist → AgentCore Runtime ARN (set in deploy env)
SPECIALIST_ARNS: dict[str, str] = {
    "sales": os.environ.get("SALES_AGENT_ARN", "arn:aws:bedrock-agentcore:us-east-1:000000000000:runtime/sales-demo"),
    "orders": os.environ.get("ORDERS_AGENT_ARN", "arn:aws:bedrock-agentcore:us-east-1:000000000000:runtime/orders-demo"),
    "support": os.environ.get("SUPPORT_AGENT_ARN", "arn:aws:bedrock-agentcore:us-east-1:000000000000:runtime/support-demo"),
    "inventory": os.environ.get("INVENTORY_AGENT_ARN", "arn:aws:bedrock-agentcore:us-east-1:000000000000:runtime/inventory-demo"),
}

MAX_SPECIALIST_HOPS = 2

app = BedrockAgentCoreApp()


def invoke_specialist(name: str, user_message: str, session_ctx: dict[str, Any]) -> str:
    """Replace with bedrock-agentcore Runtime InvokeAgentRuntime (SigV4 or JWT).

    Dry-run returns a structured stub so local py_compile / unit tests work
    without AWS credentials.
    """
    arn = SPECIALIST_ARNS[name]
    payload = {
        "specialist": name,
        "runtime_arn": arn,
        "message": user_message,
        "session": session_ctx,
        "mode": "dry-run",
    }
    return json.dumps(payload)


@tool
def route_to_sales(user_message: str) -> str:
    """Route product discovery, cart, and availability questions to the sales agent."""
    return invoke_specialist("sales", user_message, {"hop": 1})


@tool
def route_to_orders(user_message: str) -> str:
    """Route order status, cancel, modify, and shipment tracking to the orders agent."""
    return invoke_specialist("orders", user_message, {"hop": 1})


@tool
def route_to_support(user_message: str) -> str:
    """Route returns, refunds, policy FAQs, and escalations to the support agent."""
    return invoke_specialist("support", user_message, {"hop": 1})


@tool
def route_to_inventory(user_message: str) -> str:
    """Route warehouse stock and merchandising ops to the inventory agent (associates)."""
    return invoke_specialist("inventory", user_message, {"hop": 1})


@tool
def escalate_to_human(reason: str, user_message: str) -> str:
    """Hand off to a human associate when risk is high or specialists are ambiguous."""
    return json.dumps(
        {
            "status": "escalated",
            "reason": reason,
            "user_message": user_message,
            "queue": "commerce-tier2",
        }
    )


SYSTEM_PROMPT = f"""You are the eCommerce supervisor for a storefront assistant.
Classify the shopper or associate request and call EXACTLY ONE route_* tool
(or escalate_to_human). Do not invent order IDs or prices.
Never call cancel/refund/inventory write APIs yourself — specialists + Gateway Policy own writes.
If the request is ambiguous, ask one clarifying question instead of multi-routing.
Max specialist hops per turn: {MAX_SPECIALIST_HOPS}.
"""

agent = Agent(
    model=BedrockModel(model_id=MODEL_ID),
    tools=[
        route_to_sales,
        route_to_orders,
        route_to_support,
        route_to_inventory,
        escalate_to_human,
    ],
    system_prompt=SYSTEM_PROMPT,
)


@app.entrypoint
def invoke(payload: dict[str, Any]) -> dict[str, Any]:
    prompt = payload.get("prompt") or payload.get("inputText") or ""
    result = agent(prompt)
    text = str(result)
    return {"output": text, "max_specialist_hops": MAX_SPECIALIST_HOPS}


if __name__ == "__main__":
    # Local smoke: dry-run tool path without Runtime
    print(route_to_orders("Where is order ORD-1001?"))