Write AI agent skills once. Sync to 23 clients. Manage across multiple repos.
AllAgents keeps your AI tooling (skills, agents, hooks, MCP servers) in one workspace and syncs it to every client your team uses — Claude, Copilot, Cursor, Codex, Gemini, and 18 more.
# Create a workspace from a shared template
npx allagents workspace init my-workspace --from myorg/templates/nodejs
cd my-workspace
# Install plugins
npx allagents plugin install code-review@claude-plugins-official
# Sync to all configured clients
npx allagents updateNo cloning required — AllAgents fetches the workspace.yaml directly from GitHub and sets up everything.
- Configure your workspace with repos, plugins, and target clients in
workspace.yaml - Sync — AllAgents copies skills, agents, hooks, and MCP servers to each client's expected paths
- Work — every team member gets identical AI tooling via git, across any client they choose
┌─────────────────┐
│ Marketplace │ GitHub repos containing plugins
└────────┬────────┘
│
▼
┌─────────────────┐
│ AllAgents │ sync & transform
│ sync │
└────────┬────────┘
│
┌────┴────┬────────┬─────────┐
▼ ▼ ▼ ▼
.claude/ .github/ .cursor/ .agents/ client paths
Tools like npx skills and npx plugins install skills to one project for one or two clients. AllAgents manages your entire AI tooling stack — skills, agents, hooks, commands, and MCP servers — across multiple repos and all your clients, from a single declarative config.
npx skills |
npx plugins |
npx allagents |
|
|---|---|---|---|
| Config | Imperative | Imperative | Declarative (workspace.yaml) |
| Scope | Single project | Single project | Multi-repo workspace |
| Artifacts | Skills | Skills, agents, hooks, commands, MCP | Skills, agents, hooks, commands, MCP |
| Clients | 43 agents | 2 (Claude, Cursor) | 23 clients simultaneously |
| Team sharing | Each dev runs install | Each dev runs install | Git-versioned — clone and go |
| Ongoing sync | One-shot install | One-shot install | allagents update keeps everything current |
| Workspace awareness | None | None | WORKSPACE-RULES injected so AI knows all repos and skills |
| Provider resilience | Per-client | Per-client | Switch clients instantly — same tooling everywhere |
workspace:
source: ../shared-config
files:
- AGENTS.md
repositories:
- path: ../my-project
description: Main project
- path: ../my-api
description: API service
plugins:
- code-review@claude-plugins-official
- my-plugin@someuser/their-repo
clients:
- claude
- copilot
- cursor| Command | Description |
|---|---|
allagents workspace init <path> |
Create a workspace (optionally --from owner/repo) |
allagents update |
Sync all plugins to workspace |
allagents plugin install <spec> |
Install a plugin |
allagents plugin uninstall <spec> |
Remove a plugin |
allagents plugin list |
List available plugins |
allagents skills add <name> |
Add a skill from a repo |
allagents skills list |
List skills and status |
allagents workspace status |
Show workspace state |
allagents self update |
Update AllAgents CLI |
See the full CLI reference for all options.
23 AI coding assistants across two tiers:
Universal (share .agents/skills/): Copilot, Codex, OpenCode, Gemini, Amp Code, VSCode, Replit, Kimi
Provider-specific: Claude, Cursor, Factory, OpenClaw, Windsurf, Cline, Continue, Roo, Kilo, Trae, Augment, Zencoder, Junie, OpenHands, Kiro
See the client support matrix for paths, hooks, commands, and MCP support per client.
my-plugin/
├── skills/ # Reusable prompts (all clients)
│ └── debugging/
│ └── SKILL.md
├── agents/ # Agent definitions
├── commands/ # Slash commands (Claude, OpenCode)
├── hooks/ # Lifecycle hooks (Claude, Factory, Copilot)
├── .github/ # Copilot/VSCode overrides
└── .mcp.json # MCP server configs
Full documentation at allagents.dev:
- Getting Started
- Workspaces Guide
- Plugins Guide
- Agent Portability
- Client Support Matrix
- Configuration Reference
- vercel-labs/agent-skills — Universal skills for AI coding assistants
- skillpm — npm-native package manager for Agent Skills
- skills-npm — Discover agent skills shipped in npm packages
- dotagents — Unified AI agent configuration management
MIT