iterative-retrieval

Optimize context retrieval for multi-agent systems by using a multi-cycle search and evaluation loop that progressively narrows down relevant codebase information through systematic relevance scoring and query refinement.

Iterative context refinement is a 4-phase loop—Dispatch, Evaluate, Refine, Loop—used when initial context is insufficient for complex codebase exploration. You need it when subagents lack necessary codebase knowledge to perform tasks effectively.

Prevent context-limit overflows by using intelligent filtering that automatically scores file relevance and identifies missing context gaps, optimizing token usage while avoiding loading the entire repository into memory.

Yes, iterative retrieval works for complex bug fixing in unfamiliar codebases by guiding a subagent to iteratively discover relevant files, like authentication and session-management modules, through systematic relevance scoring and query refinement.

The multi-cycle search loop targets complex codebase exploration tasks where initial context is insufficient or unpredictable. It is less suitable for straightforward retrieval needs where initial context is already predictable and sufficient.