iterative-retrieval

Iterative context retrieval improves multi-agent relevance by progressively narrowing down information through a four-phase loop. It refines subagent context in stages to minimize gaps while respecting tight token budgets during large-scale discovery tasks.

Subagents miss dependencies because they start with insufficient context and cannot discover required information. An iterative retrieval loop progressively discovers relevant code and documentation, resolving missed dependencies through evaluation and refinement phases.

The best way to assemble token-efficient context is applying an iterative retrieval loop. This approach progressively narrows down relevant files across multiple refinement cycles, optimizing context assembly for large codebases within tight token limits.

To set up progressive context refinement, provide an initial task to the system. It automatically performs up to three refinement cycles using a loop to dispatch, evaluate, and refine context until multiple high-relevance results are found.

Iterative retrieval works for large-scale code discovery by progressively narrowing down relevant files. It applies a four-phase loop to evaluate and refine context, achieving multiple high-relevance results with minimal information gaps.

You should not use iterative retrieval when your workflow lacks tight token budgets or multi-agent orchestration needs. It is designed specifically for tasks where context must be refined progressively and token limits require staged evaluation.