iterative-retrieval

Iterative context retrieval uses a four-phase loop to progressively converge on high-relevance files. It evaluates relevance scores, identifies context gaps, and refines queries to prevent context bloating within three cycles.

The iterative retrieval loop uses four phases: DISPATCH, EVALUATE, REFINE, and LOOP. AI agents start with partial context and progressively acquire necessary files and terminology through this controlled retrieval cycle.

Initiate a broad retrieval query to start progressive context refinement. The system refines results through up to three cycles, evaluating relevance and highlighting gaps to return only the most relevant context files.

Iterative context refinement is designed for multi-agent workflows. It enables subagents to progressively acquire necessary files and terminology through a controlled retrieval loop, solving the context problem in distributed agent environments.

The four-phase retrieval loop is limited to three cycles to converge on relevant files. It returns only the most relevant context files and highlights remaining gaps, which may require additional manual queries for highly complex codebases.