iterative-retrieval

To provide relevant context to a subagent, use an iterative retrieval pattern that progressively gathers and evaluates candidate files based on broad queries, then refines search criteria using extracted keywords and identified context gaps.

The best way to search a codebase for complex bugs is progressive context gathering, which iteratively evaluates retrieved content, scores its relevance, and pinpoints missing information to refine subsequent file searches.

Iterative context retrieval handles missing information by identifying context gaps after evaluating initially retrieved files, then refining search criteria and performing subsequent iterations up to three cycles to find the missing data.

You need progressive context gathering for subagent communication when required context is not known upfront, ensuring deterministic tasks receive sufficient and relevant files without manually pre-selecting exact code paths.

Iterative context retrieval has a maximum limit of three cycles, iterating through broad queries, keyword extraction, and gap identification to ensure enough relevant context is collected before passing it to subagents.

Yes, you can use pattern matching to refine search criteria in multi-agent systems, extracting keywords and patterns from initially retrieved candidate files to narrow down context retrieval in subsequent iterations.