iterative-retrieval

Iterative retrieval resolves multi-agent context problems by progressively refining retrieved codebase materials across multiple cycles, evaluating and filtering files to ensure subagents receive high-relevance context without causing context explosion.

Incremental codebase discovery is achieved by starting with a broad query, evaluating retrieved materials for relevance, and iteratively refining search criteria through phased dispatch and evaluation cycles to assemble a focused context set.

The maximum limit for refining code context is a hard cap of three cycles, which balances context quality and efficiency by preventing infinite looping while allowing sufficient incremental discovery and refinement.

Iterative retrieval requires no specific dependencies or external components to manage codebase context, operating as a standalone workflow strategy for filtering and focusing on high-relevance files during multi-agent tasks.

Iterative retrieval should not be used when subagents require context that can be fully predicted upfront, as the phased discovery and refinement cycles are specifically designed for incremental discovery in unpredictable codebase or document workflows.