Optimizes codebase context gathering through a progressive 4-stage refinement loop for AI subagents.
The Iterative Context Retrieval skill implements a systematic pattern to solve the context problem in multi-agent workflows where subagents lack initial clarity on necessary project files. By employing a 4-phase cycle—Dispatch, Evaluate, Refine, and Loop—this skill enables Claude to start with broad queries and progressively narrow down to the most relevant code sections, preventing context overflow while ensuring critical dependencies and terminology are captured. It is particularly effective for large codebases where traditional 'send everything' or 'guess and check' methods fail, providing a structured approach to relevance scoring and gap identification.
