PrepGenAICerts
Courses/Introduction to Subagents/The Built-in Subagents: Explore, Plan, General-Purpose
Subagent FoundationsLesson 3 of 18

The Built-in Subagents: Explore, Plan, General-Purpose

You're Already Using Subagents

Before you ever create a custom subagent, Claude Code is already delegating to built-in ones on your behalf. Understanding them shows you the patterns to copy — and each inherits the parent conversation's permissions with additional tool restrictions.

Main ConversationExploreHaiku · read-onlyPlaninherit · read-onlyGeneral-purposeinherit · all tools

Claude Code delegates to three core built-in subagents, each with its own model and tool profile.

Explore — Fast, Read-Only Search

Explore is a fast, read-only agent optimized for searching and analyzing codebases. Claude delegates to it when it needs to understand code without changing it — keeping all those file reads out of your main context.

  • •Model: Haiku (fast, low-latency) — cheap enough to read widely.
  • •Tools: read-only (denied Write and Edit).
  • •Purpose: file discovery, code search, codebase exploration.
  • •Thoroughness levels Claude can request: quick (targeted lookup), medium (balanced), or very thorough (comprehensive).
ℹ️

A speed optimization worth knowing

Explore (and Plan) deliberately SKIP your CLAUDE.md files and the session's git status to stay fast and cheap. Every other built-in and custom subagent loads both. If a project rule must reach Explore, restate it in the delegation prompt.

Plan and General-Purpose

Plan is the research agent used during plan mode: when Claude needs to understand your codebase before presenting a plan, it delegates the research to Plan. It inherits your main model, is read-only, and — importantly — its existence helps prevent infinite nesting, because subagents cannot spawn other subagents.

General-purpose is the capable all-rounder for complex, multi-step tasks that need both exploration AND action. It inherits your main model and has access to all tools. Claude reaches for it when a task requires exploring, reasoning about results, and then making changes across multiple dependent steps.

Built-inModelToolsBest for
ExploreHaikuRead-onlyFast codebase search & discovery
PlanInheritRead-onlyResearch during plan mode
General-purposeInheritAll toolsComplex explore-and-act, multi-step work

The three core built-in subagents at a glance. (Claude Code also has helper agents like statusline-setup and claude-code-guide.)

The No-Nesting Rule

One rule shapes the entire subagent system: subagents cannot spawn their own subagents. This prevents runaway recursion. It's why Plan exists as a dedicated read-only researcher for plan mode, and why — when you need multi-level delegation — you orchestrate from the main thread rather than nesting.

⚠️

Remember this constraint

A subagent has no Agent/Task tool available to it. If your workflow seems to need a subagent that itself delegates, the answer is to chain subagents from the main conversation, or use Skills — not to nest. We'll return to this when we discuss scaling.

✨

Next

You've seen the built-ins. Next: HOW Claude actually decides which subagent to use — the description-driven delegation that powers all of this.

Key Takeaways

  • ✓Claude Code ships with built-in subagents it delegates to automatically; each inherits the parent's permissions plus extra tool restrictions.
  • ✓Explore: Haiku, read-only, for fast codebase search; Claude can request quick/medium/very-thorough levels.
  • ✓Plan: inherits the main model, read-only, used during plan mode to research before presenting a plan.
  • ✓General-purpose: inherits the main model, has all tools, for complex multi-step explore-and-act tasks.
  • ✓Explore and Plan deliberately skip CLAUDE.md and git status for speed — restate critical rules in the delegation prompt if needed.
  • ✓Core rule: subagents cannot spawn subagents (no nesting); for multi-level work, chain from the main thread or use Skills.

Check Your Understanding

Test what you learned in this lesson.

Q1.Which built-in subagent is read-only and runs on Haiku for fast codebase search?

Q2.Why do Explore and Plan skip loading CLAUDE.md and git status?

Q3.Which built-in subagent should handle a complex task needing both exploration AND code changes?

Q4.What is the 'no-nesting' rule?

Practice This Lesson

PrepGenAICerts is an independent exam-prep site. It is not affiliated with, endorsed by, or sponsored by Anthropic, PBC. Claude and Anthropic are trademarks of Anthropic, PBC. Exam details are taken from publicly available Anthropic and Pearson documentation. About PrepGenAICerts.