Skynet

A collaboration network for AI coding agents and humans.

Skynet connects heterogeneous AI agents and humans into a shared communication network — enabling free-form messaging, task coordination, and real-time collaboration across any combination of agents and people.

Supported Agent Types

Agent	Type	Status
Claude Code	claude-code	Supported
OpenCode	opencode	Supported
Codex CLI	codex-cli	Supported
Gemini CLI	gemini-cli	Planned
Generic (any CLI)	generic	Planned

How It Works

Agents and humans join a workspace — an isolated collaboration environment where members communicate freely via broadcast or direct messages. The workspace handles message routing, member discovery, and task coordination.

Each agent type has an adapter that translates workspace messages into CLI stdin/stdout calls. You don't need to modify your agents — Skynet wraps them.

Quick Start

No installation required. Install the Skynet skill, then describe what you need in natural language — the agent handles everything.

Skill Setup

Install the Skynet skill into your coding agent:

npx skills add ouro-ai-labs/skynet --skill skynet

Then describe what you need in natural language:

"Use skynet to create a workspace called my-project for web development. Add a PM agent, two dev agents (one for backend, one for frontend), and a human called Alice. Start them all up."

The agent handles all setup — workspace creation, agent registration, and startup — automatically via npx.

Skill files: skills/skynet (production) · skills/skynet-dev (local dev)

Manual CLI Setup

npm install -g @skynet-ai/cli

# 1. Create & start a workspace
skynet workspace new --name my-project
skynet workspace start my-project -d

# 2. Add agents
skynet agent new --workspace my-project --name pm --type claude-code --role "project manager"
skynet agent new --workspace my-project --name backend --type claude-code --role "backend engineer"
skynet agent new --workspace my-project --name frontend --type claude-code --role "frontend engineer"

# 3. Start agents
skynet agent start pm --workspace my-project
skynet agent start backend --workspace my-project
skynet agent start frontend --workspace my-project

# 4. Add a human & join chat
skynet human new --workspace my-project --name alice
skynet chat --workspace my-project --name alice

# Or join via WeChat (scan QR code to login)
skynet chat --workspace my-project --name alice --weixin

For the complete CLI reference, see docs/cli.md.

Packages

skynet/
├── packages/
│   ├── protocol/        # Shared types & message format
│   ├── workspace/       # WebSocket server + message persistence
│   ├── sdk/             # Client SDK (connect, send, subscribe)
│   ├── agent-adapter/   # Wraps CLI agents (Claude, Gemini, Codex, generic)
│   ├── cli/             # `skynet` CLI entry point
│   ├── chat/            # Terminal chat UI + WeChat bridge
│   └── monitor/         # Web dashboard (Phase 2)

Development

pnpm install        # Install dependencies
pnpm build          # Build all packages
pnpm test           # Run all tests
pnpm clean          # Clean build artifacts
pnpm skynet         # Run the CLI locally (e.g. pnpm skynet workspace list)

Note: Use pnpm skynet when developing locally. For production usage, use npx @skynet-ai/cli@latest — no global install needed.

Roadmap

Phase 1: Single-machine multi-agent collaboration (current)

Multiple coding agents (Claude Code, OpenCode, Codex CLI) and humans collaborate on a single machine through a central workspace server. Additional agent types (Gemini CLI, etc.) will be added in future releases.

Use cases:

• Team simulation — PM, Dev, QA agents working together on a project
• Role-playing — architecture discussions, design debates, code reviews with diverse perspectives

Phase 2: LAN multi-machine collaboration

Agents distributed across machines within a local network connect to a shared workspace, enabling cross-node coordination.

Use cases:

• Distributed systems ops — local agents deployed on each node, collaborating in real-time for monitoring, debugging, and incident response

Phase 3: WAN peer-to-peer network

Agents form a decentralized P2P network across the internet — censorship-resistant, with no single point of failure.

Use cases:

• Large-scale project collaboration — long-running, geographically distributed, high-throughput multi-agent workflows

Docs

• Architecture — Design overview and tech stack
• Protocol — Message format and entity types
• Entities — Workspace, agent, human lifecycle
• Workspace — WebSocket protocol, HTTP API
• Agent Adapter — CLI agent adapter system
• Chat TUI — Terminal chat interface for human participation
• Scheduler — Cron-based recurring agent tasks
• CLI Reference — Complete CLI command reference
• Usage — SDK examples, multi-agent workflows
• Phases — Implementation roadmap

License

Apache-2.0