Comment by rcxdude

I would not really call this a prompt injection attack, since it doesn't really hijack the agent to become malicious (something the article does discuss later on). It's more a trojan that's aimed at tricking Claude specifically.