iterative-retrieval

You can use iterative retrieval to progressively refine context for subagents. A 4-phase loop dispatches broad queries, evaluates relevance, and refines search criteria up to 3 cycles to gather necessary codebase context.

The iterative retrieval loop is a 4-phase mechanism: Dispatch for initial broad queries, Evaluate to assess relevance, Refine to update criteria, and Loop to repeat with refined criteria for a maximum of 3 cycles.

You refine context retrieval by applying an iterative loop that evaluates retrieved content and updates search criteria based on that assessment. This cycle repeats up to 3 times to ensure subagents get accurate codebase context.

Progressive context refinement is needed in multi-agent workflows where subagents start working without knowing what codebase context they require. It dynamically discovers and refines the necessary context as the task progresses.

Iterative retrieval applies to any multi-agent workflow where subagents need codebase context they cannot predict upfront. It requires no specific dependencies and uses scripts to drive the context refinement process.

The main limitation of the iterative retrieval loop is the maximum cycle constraint of 3 iterations. If the required codebase context is not found within 3 refined search cycles, the subagent may lack complete context.