iterative-retrieval

To refine incomplete context for sub-agents, use iterative retrieval to progressively narrow and enrich source material. This approach applies a 4-stage loop—DISPATCH, EVALUATE, REFINE, and LOOP—to dynamically add relevant patterns while excluding low-relevance results.

The best way to locate relevant code patterns for sub-agents is using phase-based retrieval cycles. This method evaluates initial context iteratively, applying threshold-based file selection to surface necessary terminology and documentation without overloading the agents.

Iterative retrieval works by executing a 4-stage cycle: DISPATCH to gather initial files, EVALUATE to assess relevance, REFINE to dynamically adjust queries and exclude low-value results, and LOOP to repeat until threshold-based context enrichment is achieved.

You need iterative context refinement for agents when initial context is incomplete and agents struggle to identify relevant files or patterns. It provides a structured method to progressively enrich context so sub-agents can operate effectively within large codebases.

Yes, iterative retrieval excludes low-relevance results from sub-agent context. During the REFINE stage, the system dynamically adds new keywords and patterns while actively filtering out low-relevance files through threshold-based selection criteria.

Yes, you can use threshold-based file selection to narrow context for multi-agent workflows. This process evaluates source material during each retrieval cycle, ensuring only highly relevant files are passed to sub-agents to prevent context overload.