Claude Code's Sub-Agents Explained in 5 Minutes |

Claude Code has just received a game-changing upgrade with the introduction of sub-agents. Recently, several massive updates have come to Claude Code that are flying under the radar, and they have the potential to revolutionize your development workflow. We're talking about a new feature called Sub-Agents, which could practically 10x your coding efficiency and effectiveness.

What Are Sub-Agents?

About a week ago, the team at Anthropic rolled out a powerful new feature inside Claude Code: sub-agents. This feature lets you spin up specialized, multi-agent setups directly within your terminal. These are agents tailored for specific tasks, each with its own system prompts, tool sets, and even a dedicated context window.

This is a significant deal for reducing hallucinations, improving context management, and enabling more intelligent task delegation. Think of it like deploying a team of expert agents, each one handling different parts of your development workflow. One could manage Git repositories, another could focus on debugging, while a third could specialize in documentation or testing.

For those new to it, Claude Code is Anthropic's powerful CLI tool designed to work inside your terminal. It understands your codebase, helps you write and refactor code, automates routine tasks, explains complex logic, and even handles Git workflows—all through simple, natural language prompts. Now, with sub-agents, it becomes a genius on a whole new level, acting more like a project manager that assigns the right specialist to each part of your codebase.

In short, a sub-agent has a specific purpose or area of expertise. It uses its own context window, separate from the main conversation, can be configured with specific tools it's allowed to use, and includes a custom system prompt that guides its behavior.

This overall preserves context when you're dealing with multiple tasks. It allows for specialization, where an agent can be fine-tuned for a specific task rather than having a general AI agent focus on everything. You also gain reusability; after you create a sub-agent, it can be used across different projects and shared across your team. Plus, you have flexible permissions with all these different agents.

Getting Started: Your First Sub-Agent

To get started, you'll need to ensure you have the prerequisites installed: - Node.js version 18 or above.

Once that's installed, you can install Claude Code for your operating system. For instance, on Windows, you can install it using WSL (Windows Subsystem for Linux).

After installing Claude Code, follow these steps:

Open your terminal and start the agent by running the claude command.
Set your workbench, which is the directory where you want your files to be saved.
Log in to connect your API key to Claude Code.
Once connected, type /agents in the input area. This will allow you to manage the configurations for different sub-agents.

From here, you can create a new agent, choosing whether it should be a project-level or user-level sub-agent. You can then define its purpose. It's recommended to generate the initial configuration with Claude and then customize it to fit your needs. Describe the sub-agent in detail, including when it should be used, and select the tools you want it to have access to.

The Claude Code Agent Toolkit

If you're looking for inspiration, the Claude Code Agent Toolkit is a great resource. You can discover various pre-built agents and prompts to help you create different kinds of sub-agents. For example, if you want a "UX Optimizer," you can find a prompt for it, copy it, and provide it to your new sub-agent configuration.

Step-by-Step: Creating a 'UX Optimizer' Agent

After selecting "Create New Agent," provide the location where it should operate (e.g., within the project-wide directory for a project agent).
You can either generate the configuration with Claude or set it up manually. For a manual setup, you define the agent type. For our example, this would be UX-Optimizer.
Paste the detailed agent prompt you copied from the toolkit. This will become the new description and system prompt for the sub-agent.
Next, select the tools the agent can use. The interface shows all available tools, making selection easy. You can grant it access to all tools or pick specific ones.
Finally, you can assign a background color for the agent's output to easily distinguish it in the terminal.

With that, your new sub-agent is created and ready to use within Claude Code.

Putting Sub-Agents to Work: A Practical Example

To showcase this in action, let's imagine we've used Claude Code to create a sign-up workflow. The initial version is functional but not user-friendly, with numerous input fields that create a cluttered experience.

The Problem: The sign-up form requires too much information upfront, creating a high cognitive load for the user.

The Solution: We can use the UX-Optimizer sub-agent we just created to tackle this. The goal is to redesign the workflow to reduce the process to just two or three steps while improving clarity.

Here's how the sub-agent can approach this: 1. The UX-Optimizer analyzes the existing sign-up flow. 2. It identifies non-essential fields that can be moved to a later stage of user onboarding. 3. It redesigns the flow to be more intuitive and modern, simplifying the user interface.

While the UX-Optimizer is working on the logic and flow, we could even deploy another sub-agent, perhaps a UI-Agent, to work alongside it, focusing specifically on creating a unified and aesthetically pleasing interface for the new, simplified sign-up process.

The Outcome: With these two sub-agents working in parallel, we can transform the cumbersome form into a sleek, unified sign-up workflow that is far more inviting and easier to complete. The final result is a simplified process that doesn't overwhelm the user with unnecessary fields.

Advanced Use Case: The Delegator Agent

The potential of sub-agents goes even further. One advanced example is a "Delegator" sub-agent. This agent acts as a project manager; it researches tasks, writes clear descriptions, and then dispatches other specialized, background agents to handle them.

This multi-agent setup can transform your workflow by breaking down complex projects into specialized, parallel tasks, drastically boosting efficiency and reducing your mental load. It's like having an AI project manager inside your terminal, making your coding process faster and smarter.

The new sub-agents feature in Claude Code is a powerful addition that can significantly improve your efficiency. It elevates the tool into a truly intelligent, agentic AI terminal that can deeply enhance your coding workflow. It is highly recommended to explore this feature.