iterative-retrieval

Iterative code retrieval solves unknown initial context by progressively discovering relevant files through a four-phase loop: Dispatch, Evaluate, Refine, and Loop. It dynamically adjusts search patterns and keywords based on relevance scoring.

Iterative context refinement works by dispatching broad searches, evaluating candidate files, and refining search criteria dynamically. This loop prevents sub-agents from being overwhelmed by data or lacking crucial details during codebase navigation.

The best way to search a large codebase for debugging is using dynamic query adjustment. It starts with broad search terms and progressively focuses on pertinent areas by learning from evaluated content and applying exclusions.

Yes, you can use dynamic query adjustment to narrow down search patterns. By evaluating candidate files and scoring relevance, the system iteratively refines keywords and exclusions to target specific bugs like an authentication token expiry.

You need phased context refinement when sub-agents in a multi-agent workflow lack necessary context initially. It is required when progressive discovery within a codebase is needed to prevent missing crucial details or facing information overload.

Sub-agents lack crucial details because initial context is unknown in complex multi-agent workflows. Without iterative context refinement, they cannot dynamically adjust search patterns to progressively discover pertinent code areas.