iterative-retrieval

You can refine subagent context for multi-agent codebase search by automating iterative retrieval, which progressively discovers relevant files and terms through a bounded loop to build sufficient information for task execution.

Orchestrate context-sensitive tasks using a four-phase loop of DISPATCH, EVALUATE, REFINE, and LOOP, which progressively augments context by evaluating gaps and discovering necessary information as the workflow proceeds.

The iterative retrieval pattern solves the context problem by allowing subagents to discover needed information during execution, progressively augmenting their context with relevant files and terms instead of requiring full knowledge upfront.

Yes, you can limit codebase exploration to a maximum of three cycles, ensuring bounded exploration that prevents over-searching while still allowing sufficient context refinement for the subagents.

You need iterative retrieval for task orchestration when subagents are created with limited context and cannot predict what information will be required, leading to potential missed details or wasted effort during execution.

The limitation of using a bounded loop for subagent context refinement is the strict maximum of three cycles, which may prevent discovering deeply nested information if the required context exceeds the exploration limit.