Indirect Prompt Injection in LangChain Agents via Web Content Allows Arbitrary Tool Execution | AI Breaking Wire