iterative-retrieval

Improve context retrieval for sub-agents by iteratively refining search criteria through a four-phase loop: Dispatch, Evaluate, Refine, and Loop. This dynamically adjusts search parameters based on relevance scoring to gather necessary files and patterns.

Gather codebase patterns for agentic workflows by using a phased context gathering approach. It evaluates initial broad search results, identifies missing information like specific libraries, and refines subsequent searches to pinpoint exact files.

Dynamic query adjustment for codebase analysis works by learning from initial evaluations. It adds relevant keywords and patterns while excluding irrelevant files in subsequent search cycles to refine context retrieval.

You need iterative context retrieval in multi-agent systems when sub-agents lack sufficient information about relevant files, codebase patterns, or project terminology to begin their tasks effectively without manual context provisioning.

Yes, you can use iterative retrieval to fix bugs requiring project terminology. It starts with a broad search, evaluates results to identify missing terminology or libraries, and refines the search to locate exact files needed for the task.

Iterative context refinement for sub-agents depends on accurate relevance scoring during the evaluation phase. Inaccurate gap identification or poorly defined initial search criteria can reduce the effectiveness of the dynamic query adjustment loop.