iterative-retrieval

Retrieve relevant codebase context for subagents by applying an iterative four-phase loop of DISPATCH, EVALUATE, REFINE, and LOOP. This progressively surfaces high-relevance files through quantitative scoring and dynamic query refinement.

Fix missing-context or context-too-large errors by using progressive context refinement. This method incrementally identifies relevant files and terminology, scoring relevance to guide refinement and prevent context overload.

Iterative context retrieval works by performing an initial broad retrieval, then running up to three cycles of dynamic context expansion. It quantitatively scores file relevance during the EVALUATE phase to guide the REFINE phase.

Yes, you can use progressive context refinement when spawning subagents that need codebase context. It is designed for scenarios where subagents cannot know what context they need beforehand.

The main limitation of iterative context retrieval is the maximum cycle constraint. The refinement loop runs for up to three cycles, which limits the depth of dynamic context expansion before the process must return results.

No specific dependencies are required to run iterative retrieval workflows. The system operates independently to resolve the context problem in multi-agent workflows by progressively refining context retrieval.