iterative-retrieval

Provide subagent context through a 4-phase iterative retrieval loop that progressively refines codebase information. It dispatches broad queries, evaluates relevance, refines search criteria, and repeats up to 3 cycles to ensure subagents have necessary task information.

Iterative context retrieval works through a 4-phase loop: dispatching initial broad queries for candidate files, evaluating content relevance, refining search criteria to target gaps, and repeating up to 3 cycles. This progressively refines codebase context for subagents.

You need progressive context refinement when subagents require codebase context they cannot predict upfront. It is suitable for building multi-agent workflows where context is iteratively refined to ensure subagents have sufficient information to perform tasks effectively.

Yes, you can use iterative retrieval for tasks like fixing an authentication token expiry bug. It dispatches initial queries based on patterns and keywords, then iteratively refines the search criteria up to 3 cycles to gather the necessary unpredictable codebase context.

The 4-phase retrieval loop is limited to a maximum of 3 iterative cycles for refining subagent context. It relies on initial broad queries, relevance evaluation, and query refinement, which may not surface deeply nested or undocumented codebase dependencies if patterns are too narrow.