iterative-retrieval

Codebase context retrieval in multi-agent systems fails when subagents lack predictive context, leading to missing context or context too large errors. An iterative 4-phase approach dispatches broad queries, evaluates relevance, and refines search criteria to accurately target necessary files.

To fix missing context failures in multi-agent codebase tasks, use a progressive refinement loop. This approach dispatches initial broad queries, evaluates the retrieved content, and updates search criteria over up to 3 cycles to ensure subagents receive accurate context.

Progressive context refinement works through a 4-phase loop: dispatching an initial broad query to gather candidate files, evaluating retrieved content for relevance, refining search criteria, and repeating the loop up to 3 cycles. This iteratively optimizes context for subagents.

Iterative context retrieval works for large codebases with multiple subagents by running up to 3 refinement cycles. It prevents the context too large problem by evaluating and narrowing down candidate files, ensuring subagents only receive the most relevant context for their tasks.

Use an iterative retrieval process instead of a single search query when subagents lack predictive context for complex codebase tasks. If initial broad searches return too much irrelevant data or miss critical files, the 3-cycle refinement loop evaluates and updates criteria to improve accuracy.