iterative-retrieval

To build sufficient context for sub-agents, use an iterative retrieval loop to progressively discover relevant files, terms, and patterns through broad queries, relevance scoring, and query refinement across multiple cycles.

The best way to refine context scope in a multi-agent workflow is by applying patterns, keywords, and explicit exclusions to prune low-signal results and expand high-signal search space across evaluation cycles.

Iterative retrieval works through a four-stage loop: dispatching broad queries, evaluating relevance, refining search terms, and looping, allowing multi-agent systems to progressively discover relevant files and patterns up to three cycles.

Yes, you can exclude low-signal results during the context retrieval process by applying explicit exclusion patterns, ensuring sub-agents only receive high-signal, relevant contextual material for task execution.

Sub-agents make poor decisions due to insufficient initial context, which can be fixed by iteratively retrieving and refining contextual content to progressively assemble relevant files, terms, and patterns needed for aligned actions.

Yes, the iterative context retrieval process enforces a structured loop with a maximum limit of three cycles to ensure efficient discovery and ranking of contextual material without excessive query iterations.