Introducing DefenseClaw: Automatic Threat Detection for OpenClaw
DefenseClaw scans every skill and prompt your agent receives for known malicious patterns — automatically, locally, and for free. Here's why we built it and how it works.
git clone https://github.com/clawmentorai/defenseclaw ~/.openclaw/skills/defenseclawSince the ClawHavoc attacks in February, one question keeps coming up in the OpenClaw community: how do I know what's safe to install?
SafePaste helped with manual checks. ClawMentor's mentor packages come pre-verified. But there was still no automatic, always-on defense layer running inside OpenClaw itself.
Today that changes.
What DefenseClaw does
DefenseClaw is an open-source OpenClaw skill that intercepts every incoming skill, prompt injection, and configuration update before your agent processes it. It checks against a regularly updated pattern library of known attack vectors — including ClawHavoc variants, prompt injection patterns, and exfiltration attempts.
If something matches: DefenseClaw blocks it, logs it, and tells you exactly what it found. If it's clean: nothing happens. Zero friction for legitimate packages.
How it works technically
DefenseClaw runs as a skill hook that fires before any new content is loaded into your agent's context. It uses a combination of:
- Static pattern matching against known malicious signatures
- Structural analysis (unusual permission requests, exfiltration-shaped data flows)
- Hash verification for official ClawMentor mentor packages
Everything runs locally. No network calls, no telemetry, no data leaving your machine.
Why we open-sourced it
ClawMentor is a commercial product, but the threat landscape affects everyone using OpenClaw — not just our subscribers. We built DefenseClaw as a standalone free skill because the OpenClaw community needs a shared defense layer.
If you find a pattern we missed, open a PR. The pattern library is community-maintained.
Install in 30 seconds
git clone https://github.com/clawmentorai/defenseclaw ~/.openclaw/skills/defenseclawRestart OpenClaw. That's it.
Common questions
What is DefenseClaw?+
DefenseClaw is an open-source OpenClaw skill that scans any incoming skill or prompt for known malicious patterns before it reaches your agent. It runs locally — no data leaves your machine.
Is DefenseClaw made by OpenAI or the OpenClaw team?+
No. DefenseClaw is an independent open-source project published by ClawMentor. It is not affiliated with OpenAI or the official OpenClaw project.
How is DefenseClaw different from SafePaste?+
SafePaste is a manual check — you paste a prompt and it tells you if it looks suspicious. DefenseClaw runs automatically on every skill your agent receives, including updates from ClawMentor mentor packages.
Does DefenseClaw work with ClawMentor?+
Yes. All ClawMentor mentor packages are already compatibility-checked before delivery, and DefenseClaw adds a second layer of local verification. They complement each other.
How do I install DefenseClaw?+
Run: git clone https://github.com/clawmentorai/defenseclaw ~/.openclaw/skills/defenseclaw — then restart OpenClaw. That's it. No configuration required.
Protect your OpenClaw setup
Get safety-checked configuration updates from verified mentors. Cancel anytime.
Get Ember's Package — $29/moCancel anytime · 30-second install