iterative-retrieval

Iterative retrieval progressively refines codebase context for subagents in multi-agent workflows using a 4-phase loop: DISPATCH, EVALUATE, REFINE, and LOOP. It converges on relevant files to minimize token usage while maximizing context quality.

The best way to explore large repositories with context size limits is iterative retrieval. It applies relevance scoring and gap identification across up to 3 cycles, progressively assembling high-value context for agents to solve complex codebase investigations.

The iterative retrieval loop evaluates dispatched file search results to score relevance and identify context gaps. It then refines search criteria in subsequent cycles, repeating up to 3 times to assemble high-value context for subagents.

Yes, iterative retrieval is specifically designed for bug hunts, feature explorations, and complex codebase investigations. It handles scenarios where context grows or changes across tasks by refining codebase context for subagents.

The iterative retrieval loop has a maximum limit of 3 cycles. The 4-phase DISPATCH, EVALUATE, REFINE, and LOOP sequence repeats up to 3 times to converge on relevant files and assemble high-value context.

No specific dependencies are required to run the iterative retrieval process. It operates independently to progressively refine codebase context for subagents, making it easily adoptable for multi-agent workflows.