subagent-strategy

Subagent delegation for parallel analysis works by dispatching multiple focused subagents simultaneously via the Agent tool. Each subagent receives a self-contained prompt to execute research or analysis independently, and their findings are synthesized and integrated back into the main task.

You can assign specialized agents like deep-reviewer, fast-implementer, Explore, and general-purpose to match specific task requirements. Each subagent_type runs in an isolated, self-contained prompt to prevent context cross-contamination during codebase analysis.

Use subagent delegation for thorough codebase analysis or large research tasks that would overwhelm the main thread. Delegating to focused subagents keeps the main context lean and decision-ready by isolating execution and preventing context cross-contamination.

After parallel subagents complete their isolated analysis, you integrate their findings by synthesizing each subagent's output back into the main task. This guided workflow ensures isolated results consolidate into a cohesive outcome without cluttering the main context window.

Subagent delegation requires a dispatch mechanism via the Agent tool, where you specify the subagent_type to route the task to the correct focused agent. This ensures the subtask runs in a self-contained prompt and returns results to the main thread.