How to hand side tasks to Claude Code subagents so the work runs in its own context and only a summary comes back to your session.
Dispatch a subagent when a task would flood your conversation with logs or search results you won't reread; 📝Claude Code Subagents explains what each one loads and returns.
What You'll Need
- An interactive 📝Claude Code session in the project you're working on
Steps
- Ask in plain words — Type a request such as "Use a subagent to find every call site of the payments client." Claude delegates it, and the subagent appears in the panel below your prompt while you keep working.
- Pick a built-in — Name one when it fits, such as "Use the Explore agent to map the auth module." Explore is read-only, so it searches without editing files.
- Create a reusable specialist — Ask Claude: "Create a read-only code-reviewer subagent in .claude/agents/ that uses Sonnet." Claude writes a Markdown file with a name, description, tools, and model.
- Call it by name — Type
@and pick the subagent from the typeahead, then describe the task. Claude also delegates to it on its own when a task matches its description. - Fork for context-heavy tasks — Run
/subtaskwith the task, such as/subtask draft tests for the parser changes so far. The fork inherits your whole conversation, so you don't re-explain. - Check on running work — Run
/tasksto see each subagent's status and model, attach to one, or stop it. Results land in your conversation when each finishes. - Follow up — Ask Claude to continue a finished subagent's work, such as "Ask the reviewer to also check error handling." It resumes with its full history.
What's Next?
- 📝Claude Code Dynamic Workflows — when a job needs dozens of subagents cross-checking each other
- 📝How to Message Between Claude Code Sessions — when separate sessions, not subagents, need to share findings
- 📝Claude Code Forked Sub-Agents — Quick Reference — fork mechanics and prompt-cache economics
- 📝Claude Code Multi-Agent Orchestration — compare every way to run Claude in parallel