iterative-retrieval

Refine context retrieval by iteratively dispatching broad queries, evaluating relevance, and tightening search criteria. This pattern loops up to three cycles to gather sufficient context for complex agent tasks without exceeding limits.

Agents lack sufficient context due to the agent context problem. When agents don't know what information they need upfront, iterative retrieval progressively discovers specific files and snippets without overwhelming the entire codebase.

The best way to gather targeted information is progressive context refinement. It iteratively searches files, evaluates retrieved content, and refines search criteria to ensure agents receive necessary context for complex agentic workflows.

Retrieve specific codebase files by dispatching broad queries and evaluating retrieved content. If relevance is low, refine search criteria and loop up to three cycles to discover and retrieve the exact files needed.

The main limitation is the maximum of three retrieval cycles. If sufficient context cannot be gathered within three iterations of dispatching queries and refining criteria, the progressive refinement loop terminates.