iterative-retrieval

You can refine retrieved context for subagents using a four-phase iterative retrieval loop. It guides subagents through DISPATCH, EVALUATE, REFINE, and LOOP phases to narrow down relevant files without information overload.

The best way to prevent information overload in multi-agent retrieval workflows is applying a deterministic, cycle-based protocol. This approach dispatches targeted queries, evaluates results, and refines them iteratively instead of sending everything upfront.

You identify context gaps by guiding subagents through an iterative evaluation phase. The process explicitly identifies missing information after each retrieval cycle and refines the next query to target those specific gaps.

Yes, the iterative retrieval loop supports a maximum of three cycles. The deterministic protocol automatically stops after three refinement cycles to balance precision with processing efficiency.

You need an iterative retrieval process when initial context from your repository is insufficient and precision is required. It applies to codebases, documentation, or datasets where sending all available information upfront would cause information overload.

Yes, you can use iterative retrieval to break down complex tasks in a large codebase. It guides subagents to iteratively identify relevant files, terms, and gaps until sufficient high-relevance sources are found.