loading
loading
Named, pre-configured agent personas defined in files so you can dispatch specialized helpers with their own prompt and tools.
Unlike spinning up an ad-hoc subagent at runtime, a configured subagent is a markdown file in .claude/agents/ with a system prompt, an allowed tool list, and sometimes a pinned model, giving you a reusable specialist like @tester or @api-guardian. The orchestrator delegates a task to it; it runs in its own context window and returns just its result, keeping the main thread clean. Example: a 'security-reviewer' subagent restricted to read-only tools that only audits diffs. This is how you enforce separation of duties and stop one agent's noisy context from polluting another's.
Plainly
Think of Subagents (config) as a tool from the builder toolbox. Named, pre-configured agent personas defined in files so you can dispatch specialized helpers with their own prompt and tools.
In practice
Use it when you are installing, editing, testing, reviewing, or shipping code. In practice, define the owner, input, output, and failure mode before you rely on it.