iterative-retrieval

Gather relevant context for subagents by running a four-phase iterative retrieval loop: DISPATCH, EVALUATE, REFINE, and LOOP. This cycle progressively narrows search criteria across the codebase, applying relevance scoring and gap identification to surface necessary files, terms, and patterns.

The best way to narrow search criteria is through iterative context refinement. By repeatedly evaluating and refining search results, you converge on high-relevance files and project terminology, ensuring subagents operate with the right information.

Relevance scoring during codebase search works by evaluating retrieved files and patterns in the EVALUATE phase, identifying gaps in the current context, and then refining the search parameters in the REFINE phase to progressively surface the most helpful information.

Yes, you can use iterative retrieval to discover context from codebases, documents, or project terminology. It applies to multi-agent workflows where context needs to be discovered progressively through a cycle-limited convergence loop.

The four-phase loop relies on cycle-limited convergence, meaning the iterative retrieval process is bounded. If the initial search criteria are too broad or the codebase lacks clear patterns, the subagent may not converge on enough relevant context within the allotted cycles.

No specific dependencies are required to perform iterative context refinement. The Skill provides a disciplined approach to progressively narrow search criteria, relying on the four-phase loop mechanism rather than external tools to assemble relevant context.