Refines AI context retrieval through a multi-stage feedback loop to solve the subagent context problem.
The Iterative Retrieval Skill for Claude Code addresses the context gap in multi-agent workflows by implementing a progressive four-phase retrieval cycle: Dispatch, Evaluate, Refine, and Loop. Instead of overwhelming the LLM with unnecessary data or providing insufficient information, this skill enables agents to start with broad queries and iteratively narrow down to the most relevant files based on scoring and keyword discovery. This approach ensures high-precision context retrieval for complex tasks like bug fixing and feature implementation while respecting context window limits and reducing token waste.
Key Features
01Automated keyword and pattern discovery within codebases
02Smart filtering to exclude irrelevant paths and reduce noise
03Multi-stage feedback loop for progressive context refinement
04Gap identification to specifically target missing information
05Dynamic relevance scoring (0-1 scale) for retrieved files
061 GitHub stars
Use Cases
01Discovering implementation patterns in large or unfamiliar codebases
02Precise context gathering for bug fixes with unknown origins
03Optimizing sub-agent coordination to prevent context window overflow
