Optimizes AI context retrieval through a multi-phase loop that progressively refines code search results.
The Iterative Retrieval skill addresses the 'context problem' in complex multi-agent workflows where subagents often lack the necessary codebase knowledge to perform tasks accurately. By implementing a 4-phase loop—Dispatch, Evaluate, Refine, and Loop—this skill enables Claude to start with broad queries and progressively narrow down the most relevant files based on discovered terminology and code patterns. This approach prevents token waste from oversized context while ensuring critical code segments aren't missed during deep exploration or bug-fixing tasks.
主要功能
014-Phase refinement loop: Dispatch, Evaluate, Refine, and Loop
020 GitHub stars
03Token usage optimization through targeted file inclusion
04Automated identification of context gaps for subsequent searches
05On-the-fly learning of project-specific terminology and naming conventions
06Dynamic relevance scoring on a 0-1 scale for each retrieved file
使用场景
01Performing deep codebase exploration when search terms are initially unknown
02Locating critical files for complex bug fixes in unfamiliar or large repositories
03Initializing subagents with precise context in multi-agent orchestration
