iterative-retrieval

To ensure complete context, use iterative query refinement to ask targeted follow-up questions that bridge the semantic gap between the orchestrator and sub-agents. This structured protocol guides the sub-agent to gather specific details until the orchestrator's required context level is achieved.

Iterative retrieval is a prompt engineering technique where an orchestrator evaluates a sub-agent's response and issues targeted follow-up questions to resolve semantic gaps. This structured refinement cycle ensures comprehensive context gathering before accepting the final output.

Evaluate sufficient context by using a structured checklist to objectively assess the sub-agent's response against your requirements. If semantic gaps remain, initiate up to three targeted refinement cycles to prompt the agent for missing details.

Yes, you can resume a context gathering workflow because the structured refinement protocol outputs the cycle count, gathered context, and a specific agent ID. This output format allows you to restart the session and continue iterative query refinement seamlessly.

The main limitation of iterative query refinement is the recommended maximum of three refinement cycles per sub-agent interaction. Exceeding this threshold is not advised, and complex semantic gaps may require manual orchestrator intervention rather than further automated follow-up questions.