Claude Code Subagent Generator
Describe the job in plain English, pick a specialty, and get a complete subagent file ready to save.
Question 1 of 6
What should this subagent do?
One or two sentences is plenty - be specific about the job, not the tech stack. Leave it blank and the specialty you pick next still produces a solid file.
6 quick questions - free, no sign-up, nothing leaves your browser
FAQ
What is a Claude Code subagent?
A markdown file with YAML frontmatter (name, description, tools, model, color) followed by a system prompt. Claude Code reads the description to decide when to delegate a task to it, then runs that subagent in its own separate context with only the tools listed in its frontmatter.
Where do I save the generated subagent file?
~/.claude/agents/<name>.md for a personal one that applies to every project on your machine but isn't shared, or .claude/agents/<name>.md for one committed to the repo so your whole team gets it. The generator shows the exact path for whichever you picked.
How does Claude Code decide when to use a subagent automatically?
Entirely from how the description field reads - there's no separate toggle for automatic versus manual invocation the way slash commands have. Wording like "use proactively" or naming the exact trigger condition encourages Claude to delegate on its own; a plain description without that language means it mostly waits to be asked by name.
Why do some specialties stay read-only even if I pick "can edit files"?
Code review, security audits, and docs research all stay locked to read-only tools regardless of that answer. A review or audit agent that can rewrite the code it's checking isn't providing an independent check anymore, and a docs agent's job is to read and fetch, not edit your files.
Is this Claude Code subagent generator free?
Yes - free, no sign-up, and nothing you type leaves your browser; the file is assembled locally. ClockedCode separately sells a complete done-for-you Claude Code setup with 9 subagents already built in, but the generator itself is not gated.
This generator writes a Claude Code subagent: a markdown file with YAML frontmatter that Claude delegates work to in its own separate context. Describe what you want it to do in a sentence, pick the specialty closest to the job, then answer a few more quick questions. What comes back is one file: a tools list that matches the job, a system prompt that actually says something, and the exact path to save it to.
The specialty matters more than the free-text box does. Pick code review and you get the same read-only procedure a real code-review agent follows: read the diff first, check correctness and edge cases before style, and end with a Must-fix/Should-fix/Nit list. Pick debugging and the steps change: capture the failure, form a hypothesis, confirm it before touching anything. Six specialties are built in - code review, debugging, writing tests, security audits, docs research, and prose editing - each with its own procedure and its own tools list. Pick "something else" and it builds a scaffold straight from your description instead.
Three of the six specialties stay read-only no matter what you pick for capability. Code review and security audits don't get Edit or Write, because an agent that can rewrite the code it's supposed to be auditing isn't auditing anything anymore. Docs research is read-only for a different reason: it gets WebFetch and WebSearch added automatically instead, since a research agent with no way to check a real URL is just guessing with extra steps. None of this is arbitrary - it's the same tool list an agent doing that job for real would need.
The detail people usually get wrong writing these by hand is the description field. Whether Claude reaches for a subagent on its own comes down entirely to how that field reads - there's no separate on/off switch the way slash commands have. This generator writes "Use proactively..." phrasing when you want automatic delegation, and a plain, on-request description when you'd rather type the agent's name yourself.
Once you've got one working subagent, the Claude Code agent teams guide covers running several of these together on one task, and the CLAUDE.md generator builds the instruction file that shapes how every agent behaves in between. If you're not sure a custom agent is even the right move yet, the Claude Code templates library has 9 pre-built ones you can copy first.
I run Claude Code daily and this generator writes the same tool-list and description conventions my own subagents use. If you want the finished version of the whole idea - 9 tuned subagents plus a tuned CLAUDE.md and the full vetted tool stack, installed in one paste - that is what ClockedCode is. The link is at the bottom of this page.
Other free tools
MCP Config Generator
Pick the MCP servers you use and get the exact claude mcp add commands or a ready .mcp.json block, scoped and copy-pasteable.
Try toolConfig generatorSlash Command Generator
Answer 7 quick questions and get a custom Claude Code slash command with valid frontmatter, ready to drop into .claude/commands/.
Try toolConfig generatorStatusline Generator
Toggle the segments you want, watch the exact statusline render live, then copy the tested script and settings.json block.
Try toolConfig generatorsettings.json Generator
Pick permission rules, hook presets, env vars, and a model override one step at a time, then copy or download the finished settings.json.
Try toolConfig generatorCLAUDE.md Generator
Answer 7 quick questions, get a tuned CLAUDE.md for your project in 30 seconds.
Try toolTemplate libraryClaude Code Templates
A curated library of copy-paste Claude Code templates: subagents, hooks, slash commands, settings, and CLAUDE.md starters, each one explained and pulled from a setup that gets used every day.
Try toolToken scannerUsageCut
Scan your Claude Code setup in 30 seconds and see exactly what is burning your tokens - bloated context, heavy MCP servers, and settings quietly draining your usage limits.
Try tool