Add CombinedCodeSearchTool for unified code search

[pranath-reddy]

Summary 📝

This PR adds the implementation of the CombinedCodeSearchTool, which aggregates results from multiple code search tools and reranks them using a pretrained CrossEncoder model. The combined tool queries the local semantic code search, GitHub keyword-based search (via SearxNG), and the SDE API-based search, returning a unified and ranked list of relevant code repositories.

Details

• Introduced CombinedCodeSearchTool under akd.tools.code_search to support unified querying across multiple search tools.
• Combines results from:
  • LocalRepoCodeSearchTool for vector similarity search over local repositories
  • GitHubCodeSearchTool for keyword-based GitHub repository search via SearxNG
  • SDECodeSearchTool for querying the NASA SDE code search API
• Aggregates outputs and reranks them using a pretrained CrossEncoder model (e.g., cross-encoder/ms-marco-MiniLM-L6-v2)
• All results are returned in the standard SearchResultItem format with additional metadata:
  • extra["tool"] stores the source tool name
  • extra["score"] stores the reranker score assigned by the CrossEncoder

Usage

from akd.tools.code_search import (
    CodeSearchToolInputSchema,
    CombinedCodeSearchTool,
    CombinedCodeSearchToolConfig,
    SDECodeSearchTool,
    GitHubCodeSearchTool    
)

# Configure the tool
search_cfg = CombinedCodeSearchToolConfig()

# Initialize the code search tool
search_tool = CombinedCodeSearchTool(config=search_cfg, tools=[SDECodeSearchTool(), GitHubCodeSearchTool()])

# Run the search
result = await search_tool._arun(
    CodeSearchToolInputSchema(
        queries=["landslide nepal"],
        max_results=10
    )
)

Bugfixes 🐛

• Fixed a bug in _sort_results where result["extra"] was incorrectly used to access the extra field of SearchResultItem. It is now correctly accessed as result.extra.

[NISH1001]

@pranath-reddy could you confirm if it can also accept the code search input schema?

[NISH1001]

And in the usage section could you also add different code search tools as parameters list to let know that it's the original purpose

[pranath-reddy]

@pranath-reddy could you confirm if it can also accept the code search input schema?

It doesn't work. Should I add top_k as a computed field directly to CodeSearchToolInputSchema?

[NISH1001]

@pranath-reddy could you confirm if it can also accept the code search input schema?

It doesn't work. Should I add top_k as a computed field directly to CodeSearchToolInputSchema?

Yes. Let's do it