iterative-retrieval

To refine context retrieval for subagents, implement iterative search and evaluation loops that progressively discover relevant codebase files based on identified relevance gaps. This pattern dynamically builds a precise context set during execution.

The best way to discover relevant codebase files for RAG pipelines is using progressive context refinement, which iteratively searches, evaluates, and refines search criteria to find necessary files without requiring upfront keyword knowledge.

Progressive context refinement solves the subagent context problem by iteratively searching and evaluating files, allowing subagents to dynamically discover and build their required context sets rather than relying on predefined file structures.

Yes, iterative search supports codebase exploration in complex debugging by progressively refining search criteria to discover all files related to specific features like user authentication, even when exact file names are initially unknown.

Iterative context retrieval reduces token waste in agent orchestration by evaluating file relevance and identifying gaps before inclusion, ensuring subagents only receive the precise codebase context needed for their specific task.

Dynamic context discovery is needed for multi-agent systems when subagents discover their context requirements during execution, requiring iterative searches to evaluate and refine codebase files rather than relying on static context injection.