> a matched pair of Claude Skills

Prompt injection, offence and defence.

Two companion skills for the same attack classes across tool-calling, RAG, MCP and memory-backed agents. One breaks an agent under authorised testing, one hardens it.

Offensive

AgentBreaker

A red-team playbook for LLM agents.

git clone …/agent-breakerOpen

Defensive

AgentArmor

Hardening LLM agents against prompt injection.

git clone …/agent-armorOpen

Made at Dublin AI Week, hosted by Give A Go. Built by Muhammad Murtuza Hussain.