claudeagents vs Writing Your Own Claude Code Subagents
A Claude Code subagent is just a markdown file with a short YAML header. Anyone can write one in five minutes — so why buy a kit?
Because writing one agent is easy, but writing a team of agents that are reliable, don't overlap, and work well together is slow. This page compares the two paths honestly, including when DIY is the better choice.
Quick verdict
Write your own when you need a few highly specialised agents for a unique workflow. Use claudeagents when you want a broad, tested team today and would rather customise than start from a blank file.
Write your own if…
- You need 1–5 agents for a very specific internal workflow
- Your domain is niche and no kit will match it
- You enjoy prompt engineering and iterating
- You have time to test and maintain them
Choose claudeagents if…
- You want 120+ agents working on day one
- You'd rather edit good agents than write from scratch
- You need SEO and marketing help, not just code
- You want skills and slash commands included
Side-by-side comparison
| Writing your own | claudeagents | |
|---|---|---|
| Upfront cost | Free (your time) | $49 one-time |
| Time to first useful team | Days to weeks of writing and testing | Under 60 seconds to install |
| Fit to your workflow | Perfect — you design it | Strong defaults you can edit |
| Coverage | Only what you write | Coding, SEO and marketing kits |
| Consistency | Depends on your discipline | One format and one CLAUDE.md system |
| Maintenance | On you, every Claude Code release | Updates included for life |
| Customisable | Fully | Fully — every file is plain markdown |
How a Claude Code subagent works
Each subagent is a markdown file stored in .claude/agents/ in your project, or ~/.claude/agents/ for every project. The YAML header gives it a name, a description that tells Claude Code when to use it, and optionally which tools and model it can use. The body is its system prompt.
Subagents run in their own context window, so a long code review or SEO audit doesn't crowd out your main conversation. Claude Code can pick one automatically based on its description, or you can ask for it by name.
The hidden cost of DIY
The first agent takes minutes. The tenth takes much longer, because now you are solving coordination problems:
- Writing descriptions precise enough that the right agent triggers for the right task
- Avoiding overlap between agents that do similar jobs
- Deciding which tools each agent should be allowed to use
- Testing each one on real tasks and fixing vague or wrong instructions
- Keeping everything current as Claude Code adds features
When writing your own is clearly better
If your workflow is unusual — a proprietary framework, an internal deployment process, a regulated domain — no general kit will know it. A few hand-written agents built around that knowledge will beat any generic agent. The same goes if you only ever need two or three agents: a kit would be overkill.
The best of both: start with a kit, then customise
Most teams end up somewhere in the middle. Install claudeagents for broad coverage, describe your stack once in CLAUDE.md so every agent follows it, then edit or add agents for the parts of your workflow that are truly unique. Every claudeagents file is readable markdown, so changing one is as easy as writing your own — but you start from something tested.
The bottom line
DIY is free in money but not in time. If you need a handful of niche agents, write them. If you want a complete team for building, ranking and marketing your product, start with claudeagents and spend your time customising instead of drafting.
Frequently asked questions
How long does it take to write a good Claude Code subagent?
A basic one takes minutes. A reliable one — with a precise description, sensible tool limits and testing on real tasks — usually takes an hour or more, and a full team takes days.
Can I edit the claudeagents agents?
Yes. Every agent, skill and command is a plain markdown file. Change instructions, rename agents, delete ones you don't need or add your own.
Where do Claude Code subagent files go?
Project agents go in .claude/agents/ inside your repository. Personal agents that should be available in every project go in ~/.claude/agents/.
Do I need to know prompt engineering to use claudeagents?
No. The agents work out of the box. Filling in your CLAUDE.md file is the only setup most people do.
Get 120+ agents for Claude Code
Coding, SEO and marketing kits, 54 skills and 54 slash commands. $49 one-time, lifetime updates.
See what's included →