iterative-retrieval

Iterative retrieval dynamically discovers relevant code files for subagents by using a four-phase loop: Dispatch, Evaluate, Refine, and Loop. This progressive context refinement pattern evaluates files against evolving task requirements to build a precise context set.

The best way to refine multi-agent context is using a four-phase loop: Dispatch, Evaluate, Refine, and Loop. This pattern manages context size and relevance by progressively searching and evaluating files against evolving task requirements during code exploration.

Iterative retrieval works in RAG-like pipelines by progressively searching and evaluating code files to build a relevant context set. It dynamically discovers necessary information based on evolving task requirements rather than relying on a static initial list.

Yes, you can use iterative retrieval to find files for debugging complex issues like session management. It progressively searches the codebase to find relevant files related to specific domains without requiring an exhaustive initial list.

You need progressive context refinement when subagents must discover necessary codebase information as they work. It is ideal for scenarios where the required context is unpredictable and not known upfront, such as debugging complex issues.