promptinjection

This document is an actionable attack playbook for multi-stage prompt-injection against conversational AI. It is not executable malware, but it contains detailed, operational techniques capable of causing sensitive disclosure (system prompts) or guardrail bypass in vulnerable systems. The content should be treated as high-risk guidance and restricted to controlled, authorized testing with strong defensive controls in place. No hardcoded secrets or direct exfiltration code are present in the file itself, but referenced tooling and local triggers increase the operational risk if deployed in permissive environments.