Phase 2 / Ep 11: Agent Management —— agents add, list, switch

🎯 Learning Objective: Master the creation and management of multi-Agent architectures.

1. Why Multiple Agents?

One Agent doing everything = one employee wearing ten hats. Specialized division of labor is the best practice.

graph LR
    subgraph Gateway["🦀 Gateway"]
        Router["Router"]
    end
    Router --> A1["🧑‍💼 personal\nDaily Assistant\nbasic permissions"]
    Router --> A2["💻 code-agent\nProgramming Assistant\ncoding permissions"]
    Router --> A3["📊 ops-agent\nOperations Monitoring\nfull permissions (within container)"]

2. Core Commands

# List all Agents
openclaw agents list

# Create new Agent
openclaw agents add code-agent
openclaw agents add ops-agent

# Switch default Agent
openclaw agents switch code-agent

# View Agent details
openclaw agents info code-agent

# Delete Agent
openclaw agents remove ops-agent

3. Workspace Isolation

Each Agent has an independent workspace:

~/.openclaw/agents/
├── personal/          # Default Agent
│   ├── SOUL.md
│   ├── USER.md
│   ├── TOOLS.md       # basic permissions
│   ├── memory/
│   └── sessions/
├── code-agent/        # Programming Agent
│   ├── SOUL.md        # Different persona
│   ├── TOOLS.md       # coding permissions
│   ├── skills/
│   │   └── git-assist/
│   └── sessions/
└── ops-agent/         # Operations Agent
    ├── SOUL.md
    ├── TOOLS.md       # full permissions
    └── sessions/

4. CLI vs Telegram Comparison

Next Episode Preview: Ep 12, Logging System – How to quickly pinpoint failures using logs and sessions when an Agent encounters issues.