iterative-retrieval

Optimizing context retrieval for subagents requires a multi-cycle search and evaluation loop that systematically evaluates file relevance and dynamically refines search patterns to ensure high-quality context injection.

Progressive context refinement is a 4-phase loop—Dispatch, Evaluate, Refine, and Loop—that narrows down relevant codebase information by iteratively discovering and assessing file relevance during the retrieval process.

Preventing context-limit overflows involves iteratively searching for relevant files, evaluating their relevance, and refining search criteria until the agent has the exact context needed to implement the fix.

Iterative retrieval is needed in multi-agent workflows when subagents lack sufficient information to perform tasks, which prevents both context-limit overflows and incomplete task execution.

Iterative retrieval does automatically learn project-specific naming conventions and patterns during the retrieval process through its built-in terminology discovery mechanism.

A fixed search cycle limits dynamic refinement, so the iterative retrieval approach requires systematic evaluation of file relevance and dynamic refinement of search patterns to ensure high-quality context injection.