Implements a multi-stage feedback loop to intelligently gather and refine codebase context for AI subagents.
The Iterative Retrieval skill addresses the critical 'context problem' in multi-agent workflows, where subagents often lack the necessary file awareness to perform tasks accurately. By employing a systematic four-phase cycle—Dispatch, Evaluate, Refine, and Loop—this skill enables Claude to start with broad queries and narrow down to the most relevant files based on discovered patterns and terminology. This approach prevents context window overflow and ensures that the AI operates with the highest-quality information, making it ideal for navigating complex repositories where manual context gathering is inefficient.
주요 기능
0161 GitHub stars
02Dynamic relevance scoring (0-1.0) to prioritize high-impact code files
03Gap analysis to identify missing information between retrieval cycles
04Four-phase 'Dispatch-Evaluate-Refine-Loop' architecture for context optimization
05Automated discovery of codebase-specific terminology and naming conventions
06Strict cycle management to prevent token bloat while ensuring information density
사용 사례
01Gathering comprehensive context for subagents without exceeding token limits
02Tracing complex bugs across multiple interconnected modules in unfamiliar codebases
03Scoping existing middleware and utility patterns before implementing new features
