running-adversarial-tests

MUST USE for ANY request to run an adversarial / security / pentest test against a local AI agent. Triggers on phrases like "run an adversarial test on my local agent", "run an adversarial test", "pentest my AI", "security test for my agent", "run owasp_agentic on my assistant", "test my chatbot for jailbreaks", "run a humanbound test", "run hb test on this", "test my agent". Drives the full end-to-end flow — verify humanbound MCP, detect server, discover endpoints, expose via ngrok, dispatch via humanbound MCP, then exit with a "grab a coffee" message (results land in the user's inbox by email; /humanbound-test:resume <id> is the opt-in in-band polling + findings render).