FREE SUBAGENT GENERATOR

Claude Code subagent generator: build a ready .claude/agents/ file

Describe the one job, pick a model tier and tool access, and download a valid subagent definition. No signup.

This free Claude Code subagent generator turns a task description, a model tier (sonnet, opus, haiku, or inherit), and a tool list into a ready .claude/agents/<name>.md file with correct frontmatter. It checks the description for single-responsibility scope and flags tool-list gaps (like Edit without Read) before you download. Everything runs in your browser, with no signup and no upload.

Build subagent
Tool access
Ready file
  • ✓Name "code-reviewer" is valid kebab-case.
  • ✓Task description looks scoped to a single responsibility.
  • ✓Tool list (Read, Grep, Glob) is a reasonable scope for this task.

Drop into .claude/agents/code-reviewer.md

---
name: code-reviewer
description: Use this agent when review a pull request diff for correctness bugs and flag risky patterns before merge.
tools: Read, Grep, Glob
model: sonnet
---

You are a focused subagent with a single responsibility: review a pull request diff for correctness bugs and flag risky patterns before merge.

Stay inside this scope. If the task you're given needs a different kind of work, say so instead of expanding beyond this responsibility.

FAQ

What's the difference between a subagent and the agent teams in a kit like AgenticKit?

A subagent is a single .claude/agents/<name>.md file: one name, one description Claude Code uses to decide when to auto-delegate to it, a tool list, and a system prompt. An agent team (like AgenticKit's 34-agent catalog) is a pre-built set of many subagents already scoped and wired together for a workflow. This tool builds one subagent file at a time; use it to patch a gap in an existing team or to start from scratch.

Why does it flag my task description as covering multiple responsibilities?

Subagents work best narrowly scoped: Claude Code's auto-delegation picks the right subagent by matching your request against each one's description, and a description that covers two unrelated jobs (joined by "and", "also", or several comma-separated clauses) makes that matching unreliable. If the task really is two jobs, generate two separate subagent files instead.

Why does it warn when I pick Edit or Bash without Read?

An agent that edits files or runs shell commands almost always needs to read file contents or command output first to act correctly. Selecting Edit/Write or Bash without Read is usually a sign the tool list was copied from a template rather than scoped to what this specific subagent needs.

Which model tier should I pick?

Pick haiku for simple, high-volume tasks like formatting checks; sonnet as the default for most review and implementation subagents; opus for the hardest reasoning tasks; or inherit to skip the model field entirely and let the subagent use whatever model the main session is running.

Is anything uploaded?

No. Generation and validation run entirely in your browser, and no signup is needed.

By Rakshit Yadav · Published 2026-10-31 · Last updated 2026-10-31. Next: preview which agents would run on a feature or validate a plugin manifest.

[ RUNS IN YOUR BROWSER • DROP A FILE OR TAP GENERATE • NO SIGNUP ]

Go further

Liked the free tool? Get the kits.

Engineering, Marketing, or Both: agents, skills, and slash commands you drop into any Cursor project.