iterative-retrieval

Code context retrieval minimizes token consumption by stopping the search process once sufficient high-relevance files are found. It uses relevance scoring and exclusion handling to prevent loading low-relevance paths into the sub-agent context window.

Query refinement for iterative code retrieval requires no external dependencies. It operates directly on code repositories, using internal relevance scoring and pattern detection to progressively narrow down the necessary file context.

Optimize token usage by implementing iterative retrieval that stops when sufficient high-relevance files are found. This prevents over-fetching by scoring file relevance and excluding low-value paths during the search and refinement cycles.

Using iterative retrieval for context gathering has the limitation of running a maximum of three search cycles. If the codebase is massive and highly fragmented, three iterations may not collect sufficient context, leaving the sub-agent without complete pattern coverage.