iterative-retrieval

Multi-agent workflows use iterative retrieval to dynamically gather and refine codebase context. It addresses the problem where subagents discover context requirements only after starting work, using a deterministic loop to progressively narrow the search space.

Bug-fix context assembly uses iterative retrieval to progressively refine codebase search results. By scoring relevance and updating criteria across a deterministic loop, it ensures subagents target the most relevant files for debugging.

The 4-phase loop drives iterative retrieval by sequentially executing Dispatch, Evaluate, Refine, and LOOP phases. It dynamically scores file relevance and updates search criteria across up to three cycles before relevance-based termination.

Feature implementation scoping in large repositories uses iterative retrieval to dynamically discover necessary context. It progressively narrows the codebase search space across cycles, allowing agents to locate relevant files without upfront dependency knowledge.

Iterative retrieval targets multi-agent workflows requiring emergent context during execution. It applies progressive context refinement to help subagents dynamically locate relevant codebase files as task requirements evolve.

Avoid iterative codebase retrieval for tasks requiring exhaustive file enumeration or strictly deterministic single-pass searches. It terminates after three cycles or when relevance scores peak, making it unsuitable for exhaustive repository scanning without dynamic evaluation.