iterative-retrieval

To refine subagent context, use an iterative retrieval loop that progressively narrows the search space and scores file relevance across cycles. This dispatches subagents, evaluates results, and refines the query to surface missing codebase patterns automatically.

Handling context-too-large failures in multi-agent workflows requires progressively refining the context loaded into subagents. A structured retrieval loop keeps context size manageable while preserving critical information by scoring and filtering irrelevant files.

Iterative retrieval for AI agent context works through a four-phase loop: DISPATCH, EVALUATE, REFINE, and LOOP. It cycles up to three times, using relevance scoring to extract new terminology and gaps, adapting the context loaded into subagents dynamically.

Yes, you can fix missing-context failures by applying iterative context refinement when spawning subagents. The evaluation phase identifies context gaps, and the refine phase adapts the retrieval query to extract necessary codebase patterns and files.

You need iterative retrieval for multi-agent workflows when subagents operate with limited context and struggle to determine required information. It is applied when spawning subagents that need evolving codebase context or addressing context-size limits.

Yes, iterative retrieval supports automatic adaptation to evolving codebase contexts. The refine phase extracts new patterns, terminology, and gaps across cycles, ensuring subagents receive updated and relevant file information dynamically.