AgentGuard
Real-time prompt injection and command attack detection
All Skills in This Repository (1)
Pure Emerald Level IndicatorsFrequently Asked Questions
FAQPage SchemaHow to install AgentGuard?▼
Run `npx skills add numbergroup/AgentGuard --all -g -y` in your terminal to install the skill globally.
What attacks does AgentGuard block?▼
It detects prompt injection, command injection like curl | bash or rm -rf, Unicode homoglyph tricks, Base64 obfuscation, and social engineering tactics such as fake urgency.
Does AgentGuard work with Claude and OpenClaw?▼
Yes. It runs as a Claude MCP server and as an OpenClaw skill, exposing analyze, sanitize, GitHub issue screening, and reporting tools in both.
Can AgentGuard screen GitHub issues for attacks?▼
Yes. It has specialized detection for Clinejection-style attacks hidden in issue titles and bodies, the same vector that compromised 4,000 developer machines.
Does AgentGuard send my data to external services?▼
No. All analysis runs locally using pattern-based detection with zero external dependencies, so no data ever leaves your machine.
Related Repositories in Legal & Compliance
View All in Legal & Compliance→
anthropicsclaude-for-legal
AI legal workspace for contracts, privacy, IP, litigation, and compliance
handsomestWeipatent-disclosure-skill
Draft Chinese patent disclosures and read patents in plain language
nisrulzapp-privacy-policy-generator
Generate privacy policies and terms for mobile and web apps