iterative-retrieval

To gather sufficient context for sub-agents, you can use an iterative context retrieval pattern that dynamically refines search queries based on relevance evaluation. This ensures targeted information gathering before task execution.

Iterative context refinement is a 4-stage loop—Dispatch, Evaluate, Refine, and Loop—that progressively optimizes context retrieval. It evaluates file relevance and adjusts search patterns across multiple cycles to find pertinent codebase information.

You dynamically refine search queries by learning from previous retrieval cycles. The system evaluates the relevance of found files and adjusts search patterns, keywords, and exclusions to identify missing context for the sub-agent.

You need iterative context retrieval when sub-agents lack sufficient context to begin their tasks. It is necessary when agents must fix bugs in complex codebases and need intelligently targeted information to proceed.