OpenClaw Skills Verified

中文 | English

For Human Users

What Is This?

This is a battle-tested skill sharing library between OpenClaw agents. Each skill is a workflow that some OpenClaw actually ran through on real devices, recording what environments work and what pitfalls to avoid.

Why Use It?

The Reality: Generic Knowledge vs. Real Pitfalls

Many "AI skills" online are written from generic knowledge — they look reasonable, but fall apart in practice. OpenClaw's real execution environment is complex and variable: different devices, networks, and permission configs all lead to unexpected issues.

You Can Do It Without Skills, But At a Cost

OpenClaw is smart. If you say "order me takeout," it can figure it out step by step. But this process will:

• Hit pitfalls repeatedly: ADB won't connect? Input box not recognized? Coupon didn't apply? Every step can get stuck
• Consume time: What should take 5 minutes might take 30 to debug
• Waste tokens: Repeated trial-and-error generates unnecessary API calls

The Value of This Repository

Every skill here is verified based on real execution logs:

• ✅ Reduce trial-and-error: Others' pitfalls (like "MIUI optimization breaks ADB") you don't repeat
• ✅ Quick start: Verified workflows, copy and use, no zero-to-one摸索
• ✅ Save costs: Avoid unnecessary token consumption, succeed on first try
• ✅ Continuous improvement: Team sharing, experience keeps accumulating

Is It Safe? Will It Leak Privacy?

We Enforce These Red Lines (see SECURITY.md):

• ✅ Never collect personal info (addresses, phones, names all use placeholders <YOUR_XXX>)
• ✅ Never auto-pay - All skills stop before "confirm/pay" buttons, wait for your manual click
• ✅ Don't save passwords/keys - Even if password is used, it's temporary, not logged to files
• ✅ HTTPS transmission - All API calls encrypted

Risk Notice:

• OpenClaw controls your phone/computer to execute tasks
• Sensitive operations (payment, password entry) require your confirmation
• If you don't trust a skill, don't let OpenClaw execute it

How to Start?

Method 1: Give OpenClaw a URL Copy this sentence, send to your OpenClaw:

Install discord-setup skill from https://github.com/vbgate/openclaw-skills-verified

Your OpenClaw will auto-read the repo, install the skill, then execute.

Method 2: Manual Install

./install-skill.sh integration/discord-setup en
cp -r skills/integration/discord-setup ~/clawd/skills/

Update Skills

When skills update, just re-run the install command:

./install-skill.sh integration/discord-setup en

This overwrites local version with latest.

Skill List

Skill	Purpose	Language	Region
discord-setup	Discord config and usage	🇬🇧 English / 🇨🇳 Chinese	🌍 Global
skill-contribution	Contribute skills based on real logs	🇬🇧 English / 🇨🇳 Chinese	🌍 Global
phone-control-autoglm	Phone control (AutoGLM)	🇬🇧 English / 🇨🇳 Chinese	🌍 Global
food-ordering	Meituan/Dianping food ordering	🇨🇳 Chinese	🇨🇳 China

---

For OpenClaw Agents

Install Single Skill

# Install specified skill from this repo
./install-skill.sh integration/discord-setup

# Or specify English version
./install-skill.sh integration/discord-setup en

# Or specify Chinese version (if skill supports)
./install-skill.sh integration/discord-setup zh

Usage Workflow

1. Read SKILL.md to judge if scenario matches
2. When stuck, read TROUBLESHOOTING.md for solutions
3. When needing to understand test environment, read LOG.md

Contribute Based on Real Logs (No Fabrication)

This is a hard requirement of this repo:

• ❌ No writing from imagination - Must be based on your real execution logs
• ❌ No fabricating problems and solutions - Must be pitfalls you actually encountered
• ❌ No "theoretically should work" - Must be "I tested it, result was XX"

Contribution Process:

1. Read your work logs/error records
2. Judge if there's "added value" (new device/new pitfall/new solution/old info expired)
3. Write to progressive loading structure (SOP→SKILL.md, failures→TROUBLESHOOTING.md, env differences→LOG.md)
4. Safe desensitization check (sensitive info→placeholders)
5. Report to human: "I found XX issue, suggest updating LOG.md, may I submit?"
6. After human agrees, submit

See skill-contribution skill and CONTRIBUTING.md

---

Core Principles

• One skill, one scenario - Don't mix functions
• Progressive loading - SKILL lean, details on demand
• LOG doesn't infinitely expand - Only record valuable new info
• Based on real logs - No fabrication, must test
• Security and privacy first - See SECURITY.md

---

OpenClaw team internal sharing, not official standard library.