iterative-retrieval

Context retrieval for subagents is refined through a 4-phase iterative loop that progressively narrows context payloads to high-relevance files. This staged discovery prevents subagents from receiving excessive upfront context while ensuring they retrieve the exact code files needed for their specific tasks.

The subagent context problem occurs when subagents in multi-agent workflows lack upfront context and require staged discovery to avoid excessive context payloads. Without progressive refinement, subagents either operate with insufficient context or consume too many tokens processing irrelevant code files.

The iterative retrieval loop evaluates and refines code search results through explicit evaluation scoring and refinement rules applied across four phases. The EVALUATE phase scores retrieved context for relevance, while the REFINE phase adjusts search parameters based on those scores to produce higher-relevance files.

Progressive context refinement is needed when subagents in multi-agent workflows must discover task-relevant code files without receiving excessive upfront context payloads. It applies specifically to scenarios where subagents require staged discovery and iterative evaluation to efficiently complete their assigned tasks.

Iterative retrieval improves token efficiency by narrowing context payloads to high-relevance files, but it requires multiple loop iterations to achieve optimal results. The 4-phase loop introduces processing overhead through explicit evaluation and refinement phases, which may not benefit simple tasks with minimal context requirements.