Introduction aux sous-agents

Aperçu

Learn how to use and create sub-agents in Claude Code to manage context, delegate tasks, and build workflows that keep your conversation clean and focused. Introduction to Subagents in Claude Code Subagents are specialized assistants that Claude Code can delegate tasks to, each running in its own isolated context window.

In this course, you'll learn how subagents work, how to create your own, and when to use them to keep your main conversation clean and focused. No programming experience is required — just basic comfort with a command line.

Start learning how to break complex work into focused pieces and get more out of longer Claude Code sessions. Create and Configure Custom Subagents Claude Code ships with built-in subagents for exploration and planning, but the real power comes from building your own.

You'll walk through the /agents command to create a custom subagent from scratch, configuring its description, system prompt, tool access, and model. You'll learn how each field in the config file influences when Claude delegates to the subagent and what instructions it receives.

Design Subagents That Stay Focused and Finish on Time A poorly configured subagent will wander, run too long, or produce output the main agent can't use. You'll learn four design patterns that fix this:

writing descriptions that shape input prompts, defining structured output formats as natural stopping conditions, surfacing obstacles so the main thread doesn't rediscover them, and restricting tool access to match each subagent's role.

Know When to Delegate and When to Stay on the Main Thread Not every task benefits from a subagent. You'll learn which patterns work — research, code reviews, and tasks needing custom system prompts — and which ones don't, like expert-persona prompts, sequential pipelines, and test runners that hide the output you need.

By the time you complete this course, you'll be able to create, configure, and deploy subagents that keep your context window clean, produce structured and actionable output, and make your Claude Code workflow significantly more effective.

Programme

• Introduction aux sous-agents
Découvrez comment les sous-agents maintiennent votre contexte principal propre en travaillant dans des fenêtres isolées, faites la connaissance des agents intégrés Explore et Plan de Claude Code, et lancez votre premier sous-agent personnalisé en utilisant la commande /agents.
• Conception et utilisation de sous-agents efficaces
Appliquez quatre modèles de conception pour créer des sous-agents qui restent concentrés et terminent à temps, puis apprenez quelles tâches valent la peine d'être déléguées : recherches, revues de code et invites personnalisées ; et lesquelles ne le sont pas.

---

Enseigné par

Anthropic Academy

---

Matières

Artificial Intelligence