Agent Hardening

Use this skill to audit and harden any LLM agent against adversarial attacks across messaging channels, email, MCP integrations, and web interfaces.

This is not a theoretical framework. Every rule here was earned from a real failure or a real pen test.

Use when

• setting up a new agent that will handle sensitive data
• auditing an existing agent's security posture
• hardening an agent after discovering a vulnerability
• preparing an agent for production or client-facing deployment
• reviewing channel configuration for injection resistance
• auditing MCP server connections and cross-service permissions
• evaluating tool-use permissions on any agent framework