Building open-source security tools for the AI agent era
| Project | Description | Status |
|---|---|---|
| AgentGuard | Autonomous security scanner for AI agents - OWASP ASI Top 10 + 6 novel vectors | v0.6.8 |
| MCP Scanner | Security scanner for MCP (Model Context Protocol) servers | v0.3.0 |
| AgentGuard App | GitHub App for automated PR security reviews | v0.1.0 |
| AgentGuard VS Code | VS Code extension - inline security diagnostics for agent code | v0.1.0 |
| AgentGuard Benchmark | 50 benchmark samples - 100% detection, 0% false positives | Dashboard |
| AgentGuard Demo | Live demo with GitHub Code Scanning integration | Active |
- 19 detection rules (10 OWASP ASI + 6 novel vectors + 3 standard)
- 102 automated tests
- 50 benchmark samples (100% detection rate)
- 6 major AI agent frameworks audited (3,262 findings)
- GitHub Action available on Marketplace
- dfx-agentguard on PyPI
- dfx-mcp-scanner on PyPI
- AI Agent Security (OWASP ASI Top 10)
- MCP Protocol Security
- Developer Tooling
- Novel Attack Vectors (Memory Poison, Tool Trust, Chain Amplify, Agent Collusion, Prompt Template Injection, Steganographic Injection)
Securing the autonomous web.