This kind of scenario is exactly why I haven't ever used Hermes or OpenClaw or even Grok Bot with any kind of email or communication channel. Prompt injection is a MASSIVE risk, and everyone has to keep this in mind when you connect agents to your communication pipelines