How to Create Your First Subagent in Claude Code
Build a reusable, specialized subagent with its own prompt and tool access using the /agents command.
A subagent is a specialized assistant that Claude Code can delegate work to. Each subagent has its own system prompt, its own restricted set of tools, and runs in a separate context window, so it does not clutter your main conversation. This guide creates a focused code-reviewer subagent and shows how it gets invoked.
- Claude Code installed and a project open
- A clear single job for the subagent to specialize in
- Five minutes to write a short, sharp prompt
Step 1: Open the agents manager
Inside a session, run the agents command. It opens an interactive manager where you can create new subagents, choose whether they live in the project or your user folder, and pick which tools they may use.
Step 2: Understand the file it creates
A subagent is just a Markdown file with YAML frontmatter. Project subagents live in .claude/agents, user subagents in ~/.claude/agents. The frontmatter holds the name, a description that tells Claude when to delegate, and an optional tools list. The body is the system prompt.
---
name: code-reviewer
description: Reviews recently changed code for bugs and clarity. Use proactively after edits.
tools: Read, Grep, Glob, Bash
---
You are a focused code reviewer. When invoked:
1. Run git diff to see what changed.
2. Read the changed files in full for context.
3. Report only real issues, ordered by severity.
4. For each issue give the file, the line, and a concrete fix.
Do not rewrite unrelated code. Be terse.Step 3: Invoke it
You can let Claude delegate automatically based on the description, or call the subagent explicitly by name. Explicit invocation is handy while you are testing the prompt.
Step 4: Iterate on the prompt
Open the Markdown file and tighten the instructions whenever the subagent drifts. Because it is version-controlled text, you can review changes in pull requests like any other code.
Example: a code-reviewer subagent that only reads, greps, and runs git, so it can never accidentally edit files while it reviews. The main agent stays clean while the review runs in its own context.
Watch related tutorials
12:36
22:40
7:18
18:52
32:08
28:30