iterative-retrieve

To improve context retrieval in a large codebase, you can apply a deterministic 4-phase retrieval protocol that guides subagents through dispatch, evaluate, refine, and loop phases to systematically discover relevant files.

Subagents miss relevant files during codebase search due to inconsistent context retrieval and evolving terminology. A score-based evaluation and vocabulary bridging approach resolves this by aligning file discovery with current codebase terms.

The best way to automate context retrieval for subagent operations is implementing a bounded, multi-phase search loop with clear phase definitions and relevance scoring to ensure deterministic file discovery and prevent infinite cycles.

Yes, you can use a multi-phase search protocol for codebases with evolving terminology. It applies vocabulary bridging to align outdated terms with current files, maintaining subagent accuracy during codebase context retrieval.

To prevent infinite loops during iterative context retrieval, enforce strict loop control with bounded search cycles. This ensures the subagent's dispatch, evaluate, and refine phases terminate predictably within the workflow.