iterative-retrieval

Iterative context retrieval works through a four-phase loop: Dispatch, Evaluate, Refine, and Loop. It dynamically adjusts search queries based on relevance scoring and identified information gaps to gather precise context for sub-agents.

To retrieve relevant code files for debugging, apply the phased retrieval loop to dynamically adjust search patterns and keywords. This identifies and extracts only the most relevant snippets, preventing the debugging agent from being overwhelmed.

The best way to solve the context problem in multi-agent workflows is using iterative refinement to gather context step-by-step. This prevents sending all or nothing to sub-agents, ensuring they lack no crucial information to begin tasks.

Yes, you can use dynamic querying to adjust search patterns, keywords, and exclusions during codebase navigation. The evaluation phase identifies information gaps and scores relevance to refine subsequent search queries automatically.

You need iterative refinement for information retrieval when working with complex codebases where precise context is crucial. It is applicable when sub-agents lack sufficient information to begin tasks and require dynamically adjusted searches.