CodeAgent / PythonInterpreterTool are bad at summarizing

[enzomich]

When I ask CodeAgent to summarize a long text, instead of leaving this job to the LLM it tries to whip up mediocre summarization algorithms. I worked around this issue by creating a SummarizationTool that passes the job to the LLM, but that looks convoluted. Are there more elegant ways?

def LLM_QandA(model, question):
    messages = [
        {
            'role': 'user', 
            'content': 
            [
                {
                    'type': 'text', 
                    'text': question
                }
            ]
        }
    ]
    return model(messages).content
    
class Summarization_Tool(Tool):
    name = "summarization_tool"
    description = ("This is a tool that summarizes the input (a long multi-line string)")
    inputs = {
        "long_text": {
            "type": "string", 
            "description": "The text to be summarized."
        } 
    }   
    output_type = "string"

    def __init__(self, model):
        super().__init__()
        self.model = model
        
    def forward(self, long_text):
        return LLM_QandA(self.model, r'summarize the following text:\n\n' + long_text)