iterative-retrieval

To prevent context limit overflows in agent orchestration, progressively refine retrieval queries to identify and deliver only high-relevance codebase files. This approach minimizes token waste by scoring file relevance on a 0-1 scale and excluding irrelevant content.

The best way to retrieve relevant code context for RAG pipelines is using a 4-phase iterative workflow that dispatches broad queries, evaluates relevance, refines search criteria, and loops up to 3 times. This ensures high-relevance context files while minimizing token consumption.

To find relevant files for debugging an authentication token expiry issue, dispatch an initial broad query, evaluate file relevance, refine the search criteria, and loop up to 3 times. This narrows down the codebase to deliver targeted context files.

Yes, you can use iterative retrieval to locate existing implementation patterns for new feature development. By progressively refining search criteria across up to 3 retrieval cycles, it identifies high-value context files like rate limiting patterns while excluding irrelevant content.

Iterative retrieval works with multi-agent workflows lacking upfront codebase knowledge by progressively refining context retrieval. It dispatches broad queries, evaluates file relevance on a 0-1 scale, and loops up to 3 times to deliver targeted files.

The limitation of this 4-phase iterative workflow for subagent context refinement is the maximum of 3 retrieval cycles. This cycle cap restricts the depth of search refinement, bounding the exploration scope to prevent excessive token consumption.