OpenClaw Agent Skill

The complete reference for deploying, configuring, and operating OpenClaw — the self-hosted personal AI assistant gateway.

Follows the open AgentSkills standard. Works with Kiro, Claude Code, Codex, Cursor, and any tool that supports SKILL.md.

Install

Kiro

1. Open Agent Steering & Skills panel → + → Import a skill
2. Choose GitHub → paste https://github.com/zeb-global/openclaw-skill

Or manually:

git clone https://github.com/zeb-global/openclaw-skill.git .kiro/skills/openclaw

Claude Code

git clone https://github.com/zeb-global/openclaw-skill.git .claude/skills/openclaw

OpenAI Codex

git clone https://github.com/zeb-global/openclaw-skill.git .agents/skills/openclaw

Cursor

git clone https://github.com/zeb-global/openclaw-skill.git .cursor/skills/openclaw

OpenClaw (workspace skill)

git clone https://github.com/zeb-global/openclaw-skill.git ~/.openclaw/workspace/skills/openclaw

Global install (any tool)

Replace <tool> with kiro, claude, codex, or cursor:

git clone https://github.com/zeb-global/openclaw-skill.git ~/.<tool>/skills/openclaw

What's Included

• Architecture — Gateway, channels, tools, providers
• Install & Deploy — Linux, macOS, AWS Lightsail blueprint
• Configuration — Workspace files, agents, models, channels (22 total)
• Skills System — Writing SKILL.md files, frontmatter, reference patterns
• Security — Hardening, sandboxing, DM policies, audit
• Multi-Agent — Sub-agents, channel routing, access profiles
• Automation — Cron, webhooks, hooks, standing orders
• Use Case Playbooks — Executive assistant, knowledge bot, research pipeline, helpdesk, and more

Structure

├── SKILL.md              ← Main skill file (loaded on activation)
├── references/
│   ├── automation.md     ← Cron, webhooks, hooks
│   ├── channels.md       ← All 22 channel configs
│   ├── core-concepts.md  ← Sessions, agents, models
│   ├── multi-agent.md    ← Sub-agents, access profiles
│   ├── playbooks.md      ← 8 use case playbooks
│   ├── security.md       ← Hardening, sandbox, auth
│   ├── skills-system.md  ← Allowlists, token cost, config
│   ├── tools.md          ← Built-in tools, groups, profiles
│   └── workspace-details.md ← SOUL.md, git backup
└── README.md

How It Works

The agent loads only the skill name and description at startup. When your request matches OpenClaw keywords, it loads the full SKILL.md. Reference files stay unloaded until the skill body directs the agent to read them — progressive disclosure, minimal token burn.