1System Prompt as Secret StoreA system prompt can guide model behavior, but it is not a safe place to keep passwords, tokens, policies, or other secrets.Prompt injection 2Unauthenticated Tool ChannelWhen an AI system can call tools, the path to those tools needs its own identity, integrity, and limits.Prompt injection 3Conflated ContextConflated context happens when different kinds of surrounding information are treated as one, making systems less reliable and harder to govern.Prompt injection 4Comment-to-Commit PromotionA practical guide to turning useful comments into traceable, reviewed, tested commits without losing context or control.Prompt injection 5Poisoned Knowledge BaseA poisoned knowledge base turns trusted information into a source of bad decisions, especially when artificial intelligence depends on it.Prompt injection