iterative-retrieval

Refine context retrieval for subagents by iteratively dispatching queries, evaluating retrieved content relevance, and refining search criteria up to three times. This progressive context refinement pattern ensures multi-agent workflows gather sufficient codebase context without overwhelming subagents with irrelevant information.

Subagents lack sufficient context in complex codebases because they often do not know what specific files, patterns, or terminology they need until they start working. This ambiguity is solved by progressively refining information retrieval through iterative search evaluation.

The best way to find specific code files for a subagent is using progressive context refinement. It iteratively searches and evaluates files against search criteria, looping up to three times to gather the exact code snippets required to fix bugs without retrieving irrelevant information.

Yes, you can use iterative retrieval to handle ambiguous search criteria in codebase navigation. It dispatches queries, evaluates the relevance of retrieved content, and refines the search criteria based on that evaluation, looping up to three times to resolve ambiguity.

No, progressive context retrieval does not loop indefinitely when evaluating codebase patterns. The iterative dispatch, evaluation, and refinement cycle is constrained to a maximum of three loops to gather sufficient context efficiently.

You need progressive context refinement for multi-agent workflows when a subagent requires specific code files, patterns, or terminology from a codebase to operate effectively but lacks the initial context to locate them. It is essential for complex tasks like fixing authentication token expiry bugs.