iterative-retrieval

Gathering relevant codebase context for subagents is automated through a progressive retrieval loop that refines file relevance across multiple cycles. It uses a 4-phase loop to dispatch, evaluate, and refine context, returning high-quality files while respecting token limits.

Scoping multi-file refactors with limited agent context is handled by iterative retrieval, which progressively surfaces high-relevance files. It runs a maximum of three cycles to infer file relevance, providing a deterministic fallback when initial context is insufficient.

Progressive context refinement for codebase onboarding works by applying a relevance scoring model across a 4-phase loop. It dispatches retrieval tasks, evaluates file relevance, refines the scope, and loops up to three times to ensure subagents receive sufficient context.

Yes, iterative retrieval is suitable for targeted bug fixes in unfamiliar codebases. It progressively narrows down file relevance through evaluation and refinement cycles, ensuring subagents operate with high-quality context without exceeding token limits.

When subagents start with unknown codebase context, the iterative retrieval loop provides a clear fallback by progressively inferring file relevance. It evaluates and refines context over up to three cycles, ensuring deterministic task scoping even in large repositories.

The iterative retrieval loop limits context gathering to a maximum of three refinement cycles. This constraint ensures token limits are respected, but means extremely large or highly dispersed codebases may require additional scoping if high-relevance files are not surfaced within the cycle limit.