Format of adding few shots with tool calls in system prompt

API
,
1

I want to add few shot examples that also include tool calls in api payload. Well if include that in messages as an example, it works fine but the issue is that the model always consider it as a part of conversation, If user ask what did I ask earlier, it give the vary same example even with placeholders, no matter how hard I tell in the system message that the first message is an example.

The message example:

messages.insert(
    0,
    {
        "role": "user",
        "content": "show me [PRODUCT] from [SHOP] that dont have [EXCLUDE ITEM]",
    },
)
messages.insert(
    1,
    {
        "role": "assistant",
        "content": '{"text": {"content": "Great Choice! let me search [PRODUCT] from [SHOP] excluding [EXCLUDE ITEM] for you","lang": "en","query_suggestions": ["Show me more items","Show [CHILD PRODUCTS]","Show me all items from [SHOP]"]}}',
        "tool_calls": [
            {
                "id": "call_TUKE1pSCOLRqUizQ4974BLV8",
                "type": "function",
                "function": {
                    "name": "search",
                    "arguments": '{"entity": "product","product": "[PRODUCT]","shop": "[SHOP]","excludes": "[EXCLUDE ITEM]"',
                },
            },
            {
                "id": "call_TUKE1pSCOLRqUizQ4974BLV9",
                "type": "function",
                "function": {
                    "name": "search",
                    "arguments": '{"entity": "product_category","category": "[PRODUCT CATEGORY]","lang": "en"}',
                },
            },
        ],
    },
)
messages.insert(
    2,
    {
        "role": "tool",
        "name": "search",
        "content": {
            "products": {
                "lang": "en",
                "content": '... PRODUCT CONTENT',
            }
        },
        "tool_call_id": "call_TUKE1pSCOLRqUizQ4974BLV8",
    },
)
messages.insert(
    3,
    {
        "name": "search",
        "role": "tool",
        "content": {
            "products_categories": {
                "lang": "en",
                "content": '...PRODUCT CATEGORY CONTENT',
            }
        },
        "tool_call_id": "call_TUKE1pSCOLRqUizQ4974BLV9",
    },
)
messages.insert(
    4,
    {
        "role": "assistant",
        "content": '{"text": {"content": "I found an Item that is relevant to your query. ... Product Details, Product Category Details","lang": "en","query_suggestions": ["Show me more items","Show [CHILD PRODUCTS]","Show me all items from [SHOP]"]}}',
    },
)

Now I want to include the example conversation in system prompt so the model can differentiate between system message and actual conversation. The issue is, I dont know in which format i need to design that example, since there is also JSON format instruction for response, and the model most of the time messed up when I try to include example in system prompt.

My System prompt design

"""
You:
... Chatbot Persona
Chat Instructions:
... Chat instructions

Fictitious Example of conversation betwen you and user. Do not reference it
user: show me burger from section b but i am elergic to garlic and peanuts
assistant: '{"text": {"content": "Great Choice! let me search [PRODUCT] from [SHOP] excluding [EXCLUDE ITEM] for you","lang": "en","query_suggestions": ["Show me more items","Show [CHILD PRODUCTS]","Show me all items from [SHOP]"]}}'
tool_calls: [
    {
        "type": "function",
        "function": {
            "name": "search",
            "arguments": '{"entity": "product","product": "[PRODUCT]","shop": "[SHOP]","excludes": "[EXCLUDE ITEM]"'
        },
    },
    {
        "type": "function",
        "function": {
            "name": "search",
            "arguments": '{"entity": "product_category","category": "[PRODUCT CATEGORY]","lang": "en"}'
        },
    },
]
tool:{
    "name": "search",
    "content": '{"products": {"lang": "en","content": '... PRODUCT CONTENT'}}'
}
tool:{
    "name": "search",
    "content": '{"products_categories": {"lang": "en","content": "...PRODUCT CATEGORY CONTENT"}}'
}
assistant:'{"text": {"content": "I found an Item that is relevant to your query. ... Product Details, Product Category Details","lang": "en","query_suggestions": ["Show me more items","Show [CHILD PRODUCTS]","Show me all items from [SHOP]"]}}'
"""