Getting Started with Deadfall
Deadfall detects AI agent compromise by exploiting their instruction-following behavior. When an AI agent reads a trapped file, it encounters instructions to call a verification tool — which triggers an alert.Installation
Quick Start
1. Initialize Your Project
deadfall.json— Configuration and trap registry- Updates
.gitignoreto exclude Deadfall state files
2. Create Trap Files
AI-Specific Traps (recommended for maximum detection):3. Configure MCP Client
Add the Deadfall Honey-MCP server to your MCP client:4. Start the Server
- Desktop notification
- Log entry in
deadfall_alerts.log - Webhook notification (if configured)
How It Works
Check Your Traps
Best Practices
- Use AI-specific trap types — Files like
.cursorrulesandCLAUDE.mdare automatically read by AI assistants - Deploy multiple trap types — Cover Cursor, Claude, Copilot, and MCP
- Add to every project — AI agents explore all accessible repositories
- Monitor alerts — Set up webhook delivery for real-time notifications
Next Steps
- Trap command reference — All trap types and options
- Serve command reference — Honey-MCP server and honeypot tools
- How detection works — Technical details