Bug in built-in DuckDuckGoSearchTool

[rohitrawat]

The description and output_type imply that the tool 'returns the top search results as a list of dict elements':

class DuckDuckGoSearchTool(Tool):
    name = "web_search"
    description = """Performs a duckduckgo web search based on your query (think a Google search) then returns the top search results as a list of dict elements.
    Each result has keys 'title', 'href' and 'body'."""
    inputs = {
        "query": {"type": "string", "description": "The search query to perform."}
    }
    output_type = "any"

While the forward method simply returns a semi-structured string:

    def forward(self, query: str) -> str:
        results = self.ddgs.text(query, max_results=10)
        postprocessed_results = [
            f"[{result['title']}]({result['href']})\n{result['body']}"
            for result in results
        ]
        return "## Search Results\n\n" + "\n\n".join(postprocessed_results)

This leads to generated code expecting results to be a dict when it's really a string:

results = web_search(query="largest continent")
print(results[0]['body'])

which causes this error: You're trying to subscript a string with a string index.

[aymeric-roucher]

Well spotted @rohitrawat ! Would you be interested to open a PR to fix the description, and change the output_type to "string"? Else I can do the changes

[julien-duponchelle]

I have try to submit a fix for that: #107

Not sure if it what you have in mind @aymeric-roucher because I needed to change the tool return value from a markdown string to a dictionary.

[aymeric-roucher]

I was thinking the opposite @julien-duponchelle 😄 Making it a string is easier for the LLM to process afterwards.