iterative-retrieval

Subagent context is refined using an iterative retrieval loop that progressively discovers files and patterns. The 4-phase loop dispatches, evaluates, refines, and loops to build focused context and reduce wasted exploration in multi-agent workflows.

Iterative retrieval is a 4-phase loop that progressively identifies files, terminology, and patterns for subagents. It helps subagents with limited initial context dynamically refine queries and select high-relevance information without overfetching.

Iterative context refinement is needed when addressing a bug or implementing a feature across a large codebase. It is essential for multi-agent workflows where subagents lack awareness of relevant files, patterns, or terminology.

Implement the 4-phase loop by dispatching subagents to retrieve files, evaluating relevance, refining queries for missing context, and looping until high-relevance selection criteria are met. This progressively builds a focused context for the task.

Yes, iterative retrieval works without external dependencies. It is an intermediate-level Skill that implements its own dispatch, evaluate, refine, and loop phases internally to dynamically refine subagent context.

Subagents overfetch and miss patterns due to limited initial context. The iterative retrieval loop solves this by applying high-relevance selection criteria and dynamic query refinement to progressively focus on necessary files and terminology.