iterative-retrieval

Iterative codebase context retrieval starts with a broad search, evaluates relevance scores, and progressively narrows results using query refinement. This gap-driven approach identifies missing context and adjusts keywords across up to three bounded cycles to locate the necessary files.

Multi-agent debugging in large repositories benefits from iterative context discovery. A subagent begins with broad related files, scores what it finds, then refines the search toward specific logic like session and token handling, ensuring complete retrieval without wasted tokens.

Query refinement improves code exploration by applying gap detection after an initial broad search. It identifies missing context from retrieved results and adjusts search patterns and exclusions, progressively increasing relevance scoring quality for accurate file discovery.

Yes, iterative search supports feature implementation tasks that begin with incomplete information. It applies progressive context discovery and bounded loops to evaluate and refine search results, ensuring the agent gathers the specific code context needed.

Incomplete codebase retrieval wastes tokens when agents guess at relevant files upfront. Iterative retrieval avoids this by applying relevance scoring and gap-driven refinement, systematically narrowing the search to ensure the right files appear without unnecessary exploration.