What it does
- Task decomposition: splits a broad objective into discrete workstreams that can run simultaneously
- Agent selection: routes each workstream to the most appropriate specialist based on the task type and target
- Parallel execution: runs independent workstreams at the same time, which is what makes Thorough mode fast despite its depth
- Result aggregation: collects output from all sub-agents, deduplicates findings, and synthesizes them into a unified result
- Agent lifecycle management: monitors sub-agent progress, handles failures, and re-delegates work when needed
- Cross-agent coordination: passes context between agents so recon output feeds testing agents and testing results feed verification agents