Claude Code subagents

Source: Claude Code Docs Topic: Specialized Claude Code subagents with separate context windows, scoped tools, memory, and permissions.

---

1. Subagents are isolated workers inside one session

Anthropic defines subagents as specialized assistants that each run in their own context window with their own system prompt, tool access, and permissions. They return results back to the calling session rather than becoming peer participants in the conversation.

2. Built-in and custom delegation types

Claude Code ships built-in subagents such as Explore, Plan, and general-purpose. The docs emphasize that this is not only a user customization feature; the harness itself relies on specialized delegated workers with different tool and model profiles.

3. Scoped distribution and precedence

Subagents can come from managed settings, session-only CLI JSON, project .claude/agents/, user ~/.claude/agents/, or plugins. Higher-priority scopes override lower ones. This makes subagents part of Claude Code’s broader instruction and policy layering model.

4. Fine-grained controls

The subagent frontmatter supports tool allowlists, denied tools, model choice, permission mode, hooks, MCP server scope, skills, background execution, isolation, and optional persistent memory. Plugin-delivered subagents are intentionally more restricted and cannot define certain high-risk fields like hooks or permission mode.