What problem does it solve? Subagents in multi-agent workflows start with limited context and do not know which files, patterns, or terminology a codebase uses. Sending everything exceeds context limits, sending nothing leaves agents uninformed, and guessing is often wrong. This Skill solves that context problem with progressive retrieval refinement. ## Core Features & Use Cases - Four-Phase Retrieval Loop: Dispatch broad queries, evaluate file relevance on a 0-1 scale, refine search criteria with discovered keywords and patterns, then loop up to 3 cycles. - Relevance Scoring and Gap Detection: Classify files as high, medium, low, or irrelevant, and explicitly identify missing context to drive the next refinement. - Terminology Learning: Discover codebase-specific naming conventions (e.g., "throttle" instead of "rate limit") during early cycles to improve later searches. - Use Case: When fixing an authentication token expiry bug, the loop starts with broad "token/auth" searches, discovers "refresh" and "jwt" terms, and converges on the four most relevant files within two cycles. ## Quick Start Ask the agent to retrieve context for your task using iterative retrieval: start with broad keyword searches, score each file's relevance, refine the query based on gaps, and repeat up to three cycles.
