OpenClaw Agent Skill 🦞

中文文档 | English

A production-ready Agent Skill that turns Codex, Claude Code, Antigravity, or other AI assistants into a focused OpenClaw operator.

Install once, then ask your assistant to diagnose, configure, secure, and maintain a local OpenClaw deployment with much better context.

Why This Skill

• Operationally focused — tuned for real OpenClaw maintenance, not generic Q&A
• Broad coverage — channels, model providers, Gateway ops, multi-agent routing, secrets, sandboxing, and security
• Practical outputs — optimized for step-by-step diagnosis, config edits, recovery, and verification

Best For

• Running or upgrading a self-hosted OpenClaw instance
• Adding new channels, providers, or agent routing rules
• Troubleshooting startup failures, auth errors, broken bindings, and unhealthy Gateway behavior
• Checking upgrade-era behavior changes such as Control UI proxy issues, workspace plugin trust, and short-lived pairing tokens
• Giving AI assistants a reusable OpenClaw knowledge pack instead of re-explaining docs each time

Example Prompts

• Help me diagnose my local OpenClaw setup
• Set up a Telegram bot for OpenClaw
• Why is my Gateway not responding?
• Why did Control UI stop connecting after I put it behind a reverse proxy?
• Why did my workspace plugin stop auto-loading after the upgrade?
• Why do my /pair or QR setup codes expire so quickly now?
• Lock down my OpenClaw security configuration
• Add a second agent for work with isolated workspace

Synced Release Coverage

This skill currently reflects the key stable OpenClaw changes through 2026-03-13, especially:

• v2026.3.12: dashboard-v2 Control UI refresh, shared fast-mode toggle across /fast/TUI/Control UI/ACP, provider-plugin architecture for Ollama/vLLM/SGLang, sessions_yield, Slack channelData.slack.blocks, short-lived pairing bootstrap tokens, disabled implicit workspace plugin auto-load
• v2026.3.11: browser-origin validation hardening, openclaw doctor --fix cron migration, Ollama Local / Cloud + Local onboarding, Gemini gemini-embedding-2-preview memory indexing, Discord autoArchiveDuration, ACP sessions_spawn.resumeSessionId
• v2026.3.8: backup create/verify, talk.silenceTimeoutMs, Brave llm-context, openclaw acp --provenance, gateway.remote.token
• v2026.3.7: ContextEngine plugin slot, persistent ACP channel/topic bindings, Telegram topic-level routing, SecretRef support for gateway.auth.token, messages.tts.openai.baseUrl
• v2026.3.2: openclaw config validate, broader SecretRef coverage, first-class pdf tool, inline sessions_spawn attachments
• The stable release sequence skips v2026.3.10 and v2026.3.9.

It also covers the most important February 2026 additions:

• v2026.2.26: first-class openclaw secrets workflow and openclaw agents bindings|bind|unbind
• v2026.2.25: agents.defaults.heartbeat.directPolicy and the reverted default direct-heartbeat behavior
• v2026.2.6: xAI / Grok, Voyage AI, claude-opus-4-6, and openai-codex/gpt-5.3-codex
• v2026.2.2: agents.defaults.subagents.thinking

Skill Structure

OpenClaw-Skill/
├── SKILL.md                     # Main entry (core workflows, commands, troubleshooting)
└── references/
    ├── channels.md              # 20+ channel setup guides (WhatsApp, Telegram, Discord, etc.)
    ├── gateway_ops.md           # Gateway architecture, service management, remote access
    ├── multi_agent.md           # Multi-agent routing, bindings, per-agent config
    ├── providers.md             # 20+ model providers (Anthropic, OpenAI, Ollama, etc.)
    └── security.md              # Auth, access control, hardening baseline, incident response

Total: ~1,400 lines of structured reference covering all core OpenClaw functionality.

Quick Install

If you already know which assistant you use, start here:

• Codex: git clone https://github.com/stellariums/openclaw-skill.git ~/.codex/skills/openclaw
• Claude Code: clone the repo, then create an openclaw subagent with /agents
• Antigravity: cp -r openclaw-skill ~/.gemini/antigravity/skills/openclaw

Install via Agent

If you would rather let your assistant install it for you, paste this prompt:

Please install the OpenClaw skill from https://github.com/stellariums/openclaw-skill for my current assistant environment.

Use these target locations:
- Codex: clone into ~/.codex/skills/openclaw on macOS/Linux, or %USERPROFILE%\.codex\skills\openclaw on Windows.
- Claude Code: clone into ~/.claude/openclaw-skill on macOS/Linux, or %USERPROFILE%\.claude\openclaw-skill on Windows, then create or update ~/.claude/agents/openclaw.md (or %USERPROFILE%\.claude\agents\openclaw.md on Windows) so it points to SKILL.md in that repo.
- Antigravity: copy or clone into ~/.gemini/antigravity/skills/openclaw on macOS/Linux, or %USERPROFILE%\.gemini\antigravity\skills\openclaw on Windows.

After installation, verify that SKILL.md exists in the installed location, tell me exactly which path you used, and tell me whether I need to restart the assistant.

Verify Installation

After installing, try one of these prompts:

• Help me diagnose my local OpenClaw setup
• Check whether my OpenClaw configuration has any problems
• Use the openclaw skill to troubleshoot Gateway startup

If the assistant starts using SKILL.md, the installation worked.

Installation

Download This Skill

Choose whichever is easiest:

1. Clone the repository:

git clone https://github.com/stellariums/openclaw-skill.git

2. Download a ZIP archive from GitHub:

• Open https://github.com/stellariums/openclaw-skill
• Click Code → Download ZIP
• Extract it locally

For Codex

The fastest option is to clone the repo directly into your Codex skills directory:

# macOS / Linux
git clone https://github.com/stellariums/openclaw-skill.git ~/.codex/skills/openclaw

# Windows PowerShell
git clone https://github.com/stellariums/openclaw-skill.git "$env:USERPROFILE\.codex\skills\openclaw"

Restart Codex after installing so the new skill is loaded.

If you downloaded a ZIP instead, extract it and copy the folder to ~/.codex/skills/openclaw or %USERPROFILE%\.codex\skills\openclaw.

For Claude Code

Based on Anthropic's current docs, Claude Code does not expose a native skills/ directory. The closest built-in equivalent is a personal subagent in ~/.claude/agents/.

The easiest option is to run /agents inside Claude Code and create a user-level openclaw subagent that points to this repository.

If you prefer files, a practical setup is:

# macOS / Linux
git clone https://github.com/stellariums/openclaw-skill.git ~/.claude/openclaw-skill
mkdir -p ~/.claude/agents
cat > ~/.claude/agents/openclaw.md <<'EOF'
---
name: openclaw
description: Use the OpenClaw skill repository to diagnose, configure, and troubleshoot OpenClaw.
---

Use `~/.claude/openclaw-skill/SKILL.md` as the primary workflow.
Open only the needed files under `~/.claude/openclaw-skill/references/`.
EOF

On Windows, clone the repo to %USERPROFILE%\.claude\openclaw-skill, then create %USERPROFILE%\.claude\agents\openclaw.md with the same content.

After that, you can ask Claude Code to use the openclaw subagent for OpenClaw-related work.

For Antigravity (Claude)

Copy the downloaded folder to your Antigravity skills directory. If needed, rename the folder to openclaw:

cp -r openclaw-skill ~/.gemini/antigravity/skills/openclaw

On Windows, you can also copy the folder manually to %USERPROFILE%\.gemini\antigravity\skills\openclaw.

The skill will be automatically detected and triggered when you ask about OpenClaw-related tasks.

For Other AI Assistants

You can download this repository the same way, then reuse SKILL.md and references/ in any assistant that supports skill or knowledge injection.

Usage Examples

Once installed, just ask naturally:

What You Say	What the AI Does
"Help me upgrade OpenClaw"	Runs openclaw backup create, openclaw update (or npm install -g openclaw@latest), openclaw config validate, openclaw doctor --fix, restarts Gateway, verifies
"Set up a Telegram bot"	Walks through bot creation, token setup, config, and verification
"Gateway is not responding"	Runs diagnostic command ladder: status → logs → doctor → channels probe
"Control UI stopped connecting behind my reverse proxy"	Checks browser-origin validation, proxy headers, and trusted-proxy config instead of weakening auth
"My workspace plugin stopped loading after upgrade"	Verifies v2026.3.12 plugin trust behavior and switches to an explicit trust/enable path
"Pairing or QR setup codes keep expiring"	Re-runs /pair or openclaw qr setup, explains short-lived bootstrap tokens, and avoids reusing shared credentials
"Lock down my OpenClaw security"	Runs security audit, applies hardened baseline, fixes permissions
"Add a second agent for work"	Creates agent, sets up workspace, configures bindings, restarts
"EADDRINUSE error"	Identifies port conflict, runs openclaw gateway --force or changes port

Key Commands Quick Reference

# Most Used
openclaw status                    # Overall status
openclaw gateway status            # Gateway daemon status
openclaw channels status --probe   # Channel health
openclaw config validate           # Validate config before start/restart
openclaw doctor                    # Diagnose issues
openclaw security audit            # Check security posture
openclaw backup create             # Create local backup before risky changes
openclaw --version                 # Check installed version (+ git hash when available)

# Must Check After Upgrades
openclaw gateway status --deep     # Deep scan including system services
openclaw doctor --fix              # Apply safe fixes and migrations
openclaw backup verify <archive>   # Verify backup manifest + payload
openclaw security audit --deep     # Probe live gateway security posture
openclaw agents bindings           # Show resolved agent bindings

# Common Setup Commands
openclaw configure                 # Interactive wizard
openclaw config get <path>         # Read config value
openclaw config set <path> <value> # Set config value
openclaw gateway install           # Install as system service
openclaw gateway start/stop/restart
openclaw channels add              # Add channel (wizard)
openclaw channels login            # WhatsApp QR pairing
openclaw channels list             # Show configured channels
openclaw models set <model>        # Set default model
openclaw models status --probe     # Check auth status
openclaw models auth add           # Add provider auth interactively
openclaw secrets reload            # Reload secret refs
openclaw security audit --fix      # Auto-fix issues

Documentation Source

This skill is built from the official OpenClaw Documentation, covering:

• Install
• Gateway Architecture
• Configuration
• Channels
• Model Providers
• Tools
• Multi-Agent Routing
• Security
• Troubleshooting
• CLI Reference

License

This skill is provided as-is for use with AI assistants. OpenClaw itself is MIT licensed.

Contributing

Issues and PRs welcome! If OpenClaw releases new features or changes, feel free to update the reference files accordingly.