Summary

• Adds AgentGuard to the Security & Systems category
• AI agent security framework: blocks dangerous commands, prevents data exfiltration, scans code with 24 detection rules, and manages skill trust levels
• MIT licensed, works with Claude Code, Codex CLI, Gemini CLI, and Cursor

Details

AgentGuard provides two layers of protection for AI agents:

1. Automatic hooks — block destructive commands, sensitive file access, and data exfiltration at runtime
2. Deep scan — 24 detection rules covering execution, secrets, exfiltration, obfuscation, Web3, and social engineering
3. Trust registry — capability-based access control per skill

Checklist

• Entry added alphabetically in Security & Systems
• Follows existing entry format
• Skill solves a real problem
• No duplicate entries