Purpose: Adversarial test-case catalog for prompt-injection and unsafe-agent-behavior testing, run per docs/06-ai-agents-rag/red-team-plan.md.
What belongs here: Test cases with injected payloads (fake/demo only), target skill, and expected safe behavior.
What must not be stored here: Real attack data harvested from an actual incident without sanitization; real candidate data.
Owner: [TENANT_CONFIGURATION_REQUIRED — Security / AI Governance]
Contents: prompt-injection-test-cases.md
Status: Populated — living document, extend after every new attack pattern discovered.