iterative-retrieval

Refine context retrieval for sub-agents by iteratively searching and evaluating files through a four-phase loop. This progressively narrows down relevant files based on task relevance, ensuring sub-agents receive precise actionable information without exceeding context limits.

Iterative context refinement is a four-phase loop of Dispatch, Evaluate, Refine, and Loop that progressively narrows down relevant files. It solves the context problem in multi-agent workflows where sub-agents lack sufficient information to identify relevant files or code patterns.

Sub-agents fail with broad queries because initial searches lack specific keywords. This skill handles this by adapting search criteria, dynamically building new search patterns, keywords, and exclusions based on relevance evaluation across multiple iterations.

Yes, you can use iterative search to fix authentication and session management bugs. The skill intelligently identifies the most relevant files related to these specific areas even when the initial query is too broad, adapting search patterns to find precise code patterns.

The best way to navigate a codebase without exceeding context limits is using dynamic query building. It adapts search patterns, keywords, and exclusions based on relevance evaluation, ensuring sub-agents receive only precise actionable information rather than excessive file data.

Iterative file retrieval is unnecessary when sub-agents already have specific keywords or know exactly which files are relevant. It is designed for scenarios where initial queries are too broad or lack specific keywords, requiring adaptive search criteria to find relevant code patterns.