iterative-retrieval

Gather unknown codebase context for subagents by running an iterative retrieval loop. This process progressively discovers relevant files through a 4-phase cycle of dispatching, evaluating, refining, and looping until enough context is collected.

Refine subagent context in large codebases by applying a structured 4-phase retrieval loop. It evaluates relevance scores and identifies gaps across up to three cycles to progressively discover necessary files.

Iterative retrieval determines enough context is gathered by applying gated quality checks. The mechanism requires identifying at least three high-relevance files and resolving context gaps before exiting the loop.

Iterative retrieval is designed for multi-agent workflows with dynamic context. It orchestrates progressive discovery for scenarios like debugging and codebase exploration where startup context is unknown.

A fixed cycle limit for context refinement restricts discovery to a maximum of three loops. If the codebase is excessively large and scattered, the gated requirement of three high-relevance files might not be met within this constraint.

No specific dependencies are required to implement progressive context discovery. The mechanism operates as a standalone orchestration loop using standard subagent dispatch and evaluation techniques.