Sub-Agent
A Sub-Agent is a subordinate intelligent unit derived and scheduled by an orchestrator agent within a multi-agent system. It is responsible for receiving subtasks decomposed by the orchestrator, executing specific functions within an isolated context, and returning results to the orchestrator. Its core characteristics include task specialization, context isolation, and a hierarchical collaborative relationship with the orchestrator agent. Sub-agents are commonly used to construct scalable pipelines for complex task processing.
Core Facts
Claude Code supports a sub-agent feature that can launch independent reasoning instances within the main session for code review.
In multi-agent collaborative architectures, the controller agent is responsible for task planning and scheduling, while sub-agents serve as execution units with specific tool access permissions and behavioral constraints.
Timeline (last 90 days)
In a typical multi-agent framework, after a sub-agent task ends, its conversation context is often collapsed or removed from the active view, requiring users to manually search through historical session records.
In multi-agent collaborative architectures, the controller agent is responsible for task planning and scheduling, while sub-agents serve as execution units with specific tool access permissions and behavioral constraints.
All Facts (5)
Claude Code supports a sub-agent feature that can launch independent reasoning instances within the main session for code review.
75%VerifiedIn multi-agent collaborative architectures, the controller agent is responsible for task planning and scheduling, while sub-agents serve as execution units with specific tool access permissions and behavioral constraints.
70%UnverifiedSkill也被称为Sub-Agent，是大智能体中的一个子环节，概念来源于多智能体系统（Multi-Agent System）
85%UnverifiedBy splitting tasks into independent contexts for execution, Sub-Agents give each subtask a smaller and more focused instruction set, which can significantly reduce the probability of instruction loss.
60%UnverifiedIn a typical multi-agent framework, after a sub-agent task ends, its conversation context is often collapsed or removed from the active view, requiring users to manually search through historical session records.
50%