iterative-retrieval

Fix subagent context gaps by using iterative retrieval to progressively refine codebase search queries across multiple cycles. This dispatch-evaluate-refine-loop workflow identifies missing context and returns only high-value files to complete code exploration tasks successfully.

Optimize token consumption in multi-agent orchestration by applying iterative retrieval with relevance scoring and gap identification. This filters irrelevant content and caps retrieval cycles, minimizing token usage while returning necessary codebase context for feature implementation or bug fixes.

Iterative retrieval works through a 4-phase dispatch-evaluate-refine-loop workflow for retrieving codebase context. It searches for relevant files, evaluates them against task requirements, identifies missing context like specific terminology, and refines search queries in subsequent cycles to return high-value files.

You need iterative retrieval for subagent context when required codebase information cannot be predicted prior to subagent execution. It is necessary for feature implementation tasks where spawned subagents lack context, preventing failed executions or context limit overruns.

Yes, you can use iterative retrieval to cap cycles and avoid unnecessary token usage. For example, when adding rate limiting to API endpoints, you can cap retrieval at 3 iterative cycles to prevent excessive token consumption while gathering sufficient context.

Yes, iterative retrieval supports RAG-like retrieval pipelines for codebase exploration. It applies relevance scoring and gap identification to filter irrelevant content and refine search queries, ensuring high-value context is returned for multi-agent workflows without exceeding context limits.