iterative-retrieval

Reduce token waste in multi-agent subagent workflows by applying a 4-phase iterative retrieval cycle that progressively narrows down relevant codebase context instead of sending full or guessed data upfront.

The best way to retrieve relevant source files for a bug fix is implementing an iterative dispatch-evaluate-refine-loop cycle that uses relevance scoring to pull only critical context and exclude irrelevant paths.

Subagents fail to find the right codebase context because they cannot predict exact requirements upfront, causing missing information or context overflow; terminology discovery fixes this by learning project naming conventions.

Yes, you can use iterative retrieval to optimize a RAG pipeline for codebases of any size by scoring retrieved files on a 0-1 scale to prioritize high-value content and minimize token usage.

You need progressive context refinement for agent orchestration tasks when subagents experience context limitations from missing information, token waste from guessed content, or context overflow from irrelevant data.

Yes, iterative retrieval works for code exploration by progressively refining codebase retrieval through a loop cycle that ensures only critical context is loaded, preventing the need to send the entire codebase into subagent context.