iterative-retrieval

Iterative context refinement solves the subagent context problem by progressively narrowing knowledge across up to three cycles. It applies a DISPATCH, EVALUATE, REFINE, and LOOP loop to incrementally filter broad codebase context into high-relevance files, ensuring subagents receive targeted knowledge for tasks like debugging.

Iterative retrieval narrows context for subagents by applying criteria-driven re-queries within a four-phase loop. Starting with broad context, it evaluates and refines results over up to three cycles, ensuring subagents handling large codebases receive only high-relevance files for their specific tasks.

Yes, iterative retrieval handles large codebases by progressively refining context for subagents. It applies to code exploration, debugging, and multi-agent task orchestration, using a four-phase loop with criteria-driven re-queries to narrow context and prioritize high-relevance results over up to three cycles.

Iterative context refinement runs up to three cycles to progressively narrow codebase context for subagents. It applies a four-phase loop—DISPATCH, EVALUATE, REFINE, LOOP—using criteria-driven re-queries to prioritize high-relevance files and exclude low-signal data from the knowledge provided.

Avoid using iterative retrieval when the codebase is small or the required context is already fully known. The pattern is designed for large codebases or partially unknown context where progressive refinement across multiple cycles is necessary to exclude low-signal files and prioritize high-relevance results.