Here’s the short answer: you install pre-built agents from Cloud Code Templates, you make Claude check that code for anything malicious before it ever touches your machine, you set the agent to global so it works across every project, and then you turn on agent teams so multiple agents can build, document, and review your project at the same time. That’s the whole system.
Want to go deeper? I run a live Claude Code workshop — check the latest dates and see if you can still grab a seat.
Disclosure: some links in this post are affiliate links. If you sign up through them I may earn a commission, at no extra cost to you.
I teach GoHighLevel and AI as one connected system, and using Claude Code alongside GoHighLevel is exactly the kind of thing I mean. Claude Code on its own is powerful, but if you’re doing everything in one long chat, it eventually falls apart. The fix isn’t more prompting. It’s agents — specialized workers with their own job, their own memory, and (eventually) their own teammates — that you can point at your GoHighLevel builds and other projects alike.
What is a sub agent in Claude Code?
A sub agent is a specialized version of Claude that’s been given one job — front-end development, documentation, copywriting, whatever you need — and its own separate memory. When you call it, it does its task and reports back to your main session instead of dumping everything into your one shared context.
That matters more than it sounds like it should. Each agent keeps a confined memory instead of re-reading your entire conversation history every time. Lower token costs, cleaner output — the agent isn’t dragging around context it doesn’t need.
How do you install a global agent from Claud Code Templates?
Cloud Code Templates is one of several sites that host ready-made agents, skills, commands, and hooks. Inside it you’ll find a library of agents sorted by popularity — start there, because a heavily downloaded agent has already been tested by a lot of other people.
The install itself is simple: click the agent, copy the install command, paste it into your terminal inside VS Code. But there’s a step before that you should never skip.
Security-check the code before you install anything
Never install code onto your computer blindly, no matter how trusted the source looks. Before installing anything from Cloud Code Templates, view the code (not just the overview), copy it, and paste it into a Claude chat with a direct instruction: check this for malicious code, prompt injections, or anything that could harm your machine or your service.
Here’s exactly what I tell Claude in the video:
“Please check this code for any malicious code, any prompt injections, or anything that can harm me or my computer, or my service that I’m trying to provide.”
If Claude comes back clean, you install. If it flags anything, you don’t touch it. That’s the whole rule.
Making the agent global instead of project-only
By default, a freshly installed agent may only exist for the project you installed it in. To use it everywhere, ask Claude directly to confirm it’s global — something like, “I just installed this agent, please find it and make sure it’s set to global so I can use it in all my projects.” You can verify it worked by typing forward-slash, then “agents,” and checking that it’s listed under your global agents rather than under a specific project.
On a PC, the agents themselves live in your user folder under a Claude directory — worth knowing so you can move a misplaced file if the install puts it one folder off, which does happen.
How do you build a custom agent from scratch?
You build a custom agent by starting a new terminal window, typing “agent,” choosing to create a new one, and picking manual configuration so you control every field yourself. This is where Claude Code stops being a one-size-fits-all tool and starts working the way you specifically need it to — including agents built around GoHighLevel workflows.
Start a new terminal window, type “agent,” and choose to create a new one. Pick “personal” instead of “project” to make it global from the start, and choose manual configuration so you understand every field instead of letting Claude auto-generate it.
There’s also a hybrid move worth knowing: take a solid pre-built agent (like a copywriting agent from Cloud Code Templates) as a starting point, then ask Claude to rewrite it around your own style — based on everything it already knows from content you’ve written together. You give it a unique name so it doesn’t collide with anything else, then paste in, in order: the identifier, the system prompt, when the agent should be used, which tools it can access, and which model it should run on (Sonnet for creative or complex work, Haiku to keep costs down, or “inherit” to match the parent model). Give it a color, turn on agent memory, and it’s live.
How do sub agents run in parallel across projects?
Once an agent exists, you can call it directly — open a project, start Claude Code in the terminal, type the agent’s name, and give it the task. You can also just describe what you need in plain language, and Claude will invoke the right agent on its own if it recognizes the task matches that agent’s description.
The real payoff shows up when you’re running more than one project. Open a second window for a different project, invoke a different sub agent there, and both run independently at the same time — each with its own context window, each reporting back to its own session. That’s what parallel actually means here: not one agent doing two things, but multiple agents each doing one thing, at once.
What are agent teams, and are they worth the extra cost?
Agent teams are the experimental next step up from independent sub agents, letting agents talk to each other directly instead of reporting back separately. A front-end agent and a backend agent can see what the other is building in real time, cutting down on conflicts your main session would otherwise have to untangle — at a meaningfully higher token cost.
Turning it on means updating Claude Code to at least version 2.1.32, then enabling the experimental agent-teams flag in your settings — which you can do by asking Claude directly to enable it globally rather than hand-editing the JSON file yourself.
Once it’s running, you can say something like “create an agent team to review my project,” and Claude will spin up a team lead plus specialized reviewers — front-end, backend, quality and security — all working in parallel, visible if you hit Control+T to see the task list.
The honest gotcha
Agent teams cost more — potentially five to seven times more in tokens — because the agents are constantly talking to each other instead of working in silence. I’ll be direct about this one:
“Be aware that token costs can range to five to seven times higher if the agents start talking to each other.”
If you’re on a metered plan, that adds up fast. It’s worth it when you need a full build done quickly and correctly. It’s overkill for a small task a single sub agent could’ve handled alone. Match the tool to the job — don’t reach for a full agent team out of habit.
Frequently asked questions
Do I need to know how to code to use Claude Code sub agents?
No. Every step in this process — installing, security-checking, making an agent global, even building one from scratch — is done by describing what you want to Claude in plain language and letting it handle the technical file creation.
Is it safe to install agents from Cloud Code Templates?
It can be, but never install blindly. Always pull the actual code, paste it into a Claude chat, and ask it to check specifically for malicious code and prompt injections before you run the install command.
What’s the difference between a sub agent and an agent team?
A sub agent works independently in its own context window and reports back when it’s done. An agent team is a group of agents that talk to each other in real time while working — faster results, but noticeably higher token cost.
Why does documentation matter when building agents and projects?
As projects grow or involve other people, clear documentation — a README, a how-to guide, setup notes — becomes essential. A documentation agent can generate this automatically once it’s installed and set to global.
This kind of system thinking — one connected ecosystem instead of a pile of disconnected tools — is exactly what we cover in the GoHighLevel Masterclass. If you’re new here, take a look at what Automated Marketer is, and if you’ve got questions about where GoHighLevel and AI tools like Claude fit together, check the FAQ.