iterative-retrieval

To refine context for multi-agent workflows, use an iterative retrieval pattern that dynamically identifies and evaluates relevant code files based on evolving task requirements. This progressively gathers necessary codebase patterns, terminology, and focus areas for subagents.

The best way to retrieve relevant code files without overwhelming agents is progressive context refinement. By iteratively searching and evaluating files, this approach intelligently provides only the targeted information needed, avoiding the injection of the entire codebase.

Iterative retrieval solves the subagent context problem by employing a four-phase loop: Dispatch, Evaluate, Refine, and Loop. This mechanism dynamically identifies relevant code files and evaluates them against evolving task requirements.

Yes, you can use iterative retrieval to find specific codebase patterns for debugging. For example, when debugging an authentication issue, it intelligently finds and provides only the relevant code files related to tokens, sessions, and user management.

Subagents lack sufficient context because they do not know what information they need upfront. This issue is solved by progressively refining the information provided to them through an iterative retrieval loop that evaluates files based on evolving task requirements.

You need progressive context refinement when your subagents lack sufficient context to perform tasks effectively. It is specifically required when agents cannot determine the necessary codebase patterns, terminology, and focus areas upfront in multi-agent workflows.