Soul-Md

Google DeepMind’s new research framework maps six categories of “AI Agent Traps” — adversarial techniques embedded in the environment that can hijack autonomous agents without the user or the agent knowing. With content injection attacks succeeding in up to 86% of tested scenarios, this isn’t theoretical risk. This guide walks through each of the six trap categories and gives you concrete, actionable mitigations you can implement today — whether you’re running OpenClaw, a custom LangGraph pipeline, or any other agent framework. ...

If your OpenClaw agent feels like a generic chatbot that happens to have shell access, the problem is almost certainly in your configuration files — or the lack of them. Two files, SOUL.md and HEARTBEAT.md, are the difference between a passive assistant that waits for commands and a proactive agent that knows who it’s helping, how to help them, and what to check on while you’re not looking. This guide walks through both. ...