Remember when Neo got Kung Fu uploaded into his brain? That's what this does for your AI agents. You pick the skills. You build the plugin. Your agent wakes up knowing everything you gave it.
A backend agent needs deploy scripts. A frontend agent needs design tools. A research agent needs web search. One-size-fits-all doesn't work. The Plugin Builder lets you assemble exactly the right capabilities for each agent you run.
Same builder. Different agents. Different superpowers.
You choose what goes in. Your agent gets exactly what it needs.
A skill is a capability you load into your agent. It's a markdown file that tells Claude Code what it can do and how to do it. No SDK. No framework. Just instructions.
Send and receive messages between agents. Coordinate work across your entire team of AI agents.
Create, rename, hibernate, and wake up other agents. Your agents can manage other agents.
Search through past conversations and remember context. Your agent never forgets what you discussed.
Understand how your codebase is connected. Query relationships between functions, classes, and files.
Search auto-generated documentation. Your agent checks function signatures and API specs before coding.
Break down complex tasks and track progress with persistent files. Never lose track of multi-step work.
These are the default skills. But you're not stuck with this list.
Add skills from anyone on GitHub. Remove what you don't need. Create your own.
Four steps. Five minutes. Your agent gets superpowers.
The Plugin Builder lives on GitHub. Fork it, and you have your own copy to customize.
Edit one file: plugin.manifest.json. Add skills from GitHub, remove what you don't need, mix in your own. This file is your recipe.
{
"sources": [
{ "name": "core", "type": "local", "path": "./src" },
{ "name": "code-review", "type": "git",
"repo": "github.com/alice/claude-skills.git" }
]
}
Run the build command, or just push to GitHub. CI builds automatically and commits the result back to your repo.
Install the plugin into your agent. Two ways to do it:
Local install (quick testing from your clone):
Remote install (permanent, from your fork on GitHub):
/plugin marketplace add YOUR-USERNAME/ai-maestro-plugins /plugin install ai-maestro@your-marketplace
Your fork, your marketplace. Because you're installing from your own GitHub repo, your custom plugin shows up as a separate entry in the marketplace — no conflicts with the official AI Maestro plugin. You can even have both installed side by side if your skills don't overlap.
A skill is just a markdown file with instructions for your agent. No SDK. No framework. Just words that tell your agent what it's capable of.
# src/skills/my-skill/SKILL.md --- name: deploy-to-production description: Deploy code safely allowed-tools: Bash, Read, Write --- # Deploy to Production When asked to deploy, run the deployment checklist: 1. Run all tests 2. Check for breaking changes 3. Build the production bundle 4. Deploy via our CI pipeline 5. Verify health checks pass
Write what you want your agent to know. Put it in a folder. Push. CI builds your plugin with the new skill included.
The Plugin Builder is one piece of the puzzle. AI Maestro is a full platform for orchestrating AI agents — the dashboard where you manage them, the messaging system where they communicate, and the skill system that gives them capabilities.
The command center. Manage all your agents from one place. Terminal streaming, team meetings, kanban boards.
Learn more →Deep dive into how skills work. Memory search, code graph, docs search — the proactive intelligence layer.
Explore skills →Agents talk to each other. Secure, signed messages. Works locally, federates globally. Slack and email integration.
See messaging →Fork the repo. Pick your skills. Five minutes from now, your agent has superpowers.