TOPIC / RESOURCE
OpenClaw glossary: Agent, Skill & core terms
Align on Agent, Skill, Workspace, Binding, Router, Gateway, and 10+ more terms before reading config or examples. Avoid common misunderstandings early.
Best for
- 看配置、案例和路由时总觉得术语没对齐
- 想先把 Agent、Skill、Workspace、Binding 这些词读顺
- 准备继续看模型、方案或案例页
Core concepts
| Term | Definition | Relevant command |
|---|---|---|
| Agent | A unit of work with its own workspace, session, and behavior rules. | openclaw config list |
| Skill | Gives the Agent a specific capability. Without skills it understands but can't act. | openclaw skills install @clawhub/xxx |
| Workspace | Where long-term rules and work files live. | — |
| Binding | Rules that route inbound messages or channels to a given Agent. | openclaw channels add |
| Fallback | Backup model when the primary fails, is unavailable, or unsuitable. | openclaw config set model.fallback |
| Router | Strategy to send different tasks to different models. Not just fallback. | See Model routing |
| Gateway | Entry layer for messages and requests into OpenClaw. | openclaw gateway status, openclaw gateway restart |
| Pairing | Binding OpenClaw to a chat channel, platform, or bot. | openclaw channels add |
Platform terms
| Term | Definition | Relevant command |
|---|---|---|
| Channel | Message entry and exit point. 12 supported: WhatsApp, Telegram, Discord, Slack, Feishu, DingTalk, Signal, iMessage, Teams, Google Chat, Matrix, WebChat. | openclaw channels add |
| Plugin | Official integration for a channel. Example: @openclaw/feishu for Feishu. | openclaw plugins install |
| ClawHub | OpenClaw skill registry with 5,700+ skills. Verified badges mean officially reviewed. | openclaw skills list |
| Sandbox | Second layer of the security model: tool sandbox filters (CPU/memory limits, read-only mounts). | — |
| Control UI / Dashboard | Local management UI, default http://127.0.0.1:18789/. | openclaw dashboard |
| DM Pairing | Pairing via DMs instead of public webhooks. Use with Gateway auth. | — |
Config and files
| Term | Definition | Relevant command |
|---|---|---|
| AGENTS.md | Agent behavior and role definition in workspace root. Persisted system prompt. | — |
| SOUL.md | Agent personality and reply style. Complements AGENTS.md. | — |
| Gateway Auth | Gateway authentication. Token mode: openclaw config set gateway.auth.mode token. | — |
| Security Audit | Built-in security check. | openclaw security audit --deep |
Common misconceptions
- Agent vs Skill: Agent is the unit; Skill adds capability.
- Workspace vs plain folder: Workspace holds long-term rules.
- Fallback vs Router: Fallback is backup; Router assigns proactively.
- Channel vs Plugin: Channel is the endpoint; Plugin adds deeper integration.
- Sandbox vs Docker: Sandbox is tool filtering; Docker is network isolation.
Next steps
Config templates · Model routing · Security
External links
FAQ
Topic FAQ
Many config and example issues come from misaligned terms, not from deep concepts.
RELATED
Related posts
Featured
What is OpenClaw? More than a chat shell
OpenClaw makes more sense as an assistant platform with channels and skills than as a simple chat wrapper.
OverviewGetting started
Read next →
Featured
OpenClaw Install Guide: Verify Before You Copy
The fastest install guide starts with the right decisions: current requirements, environment choice, and a clear first validation loop.
Install guideSetup
Read next →