Recently, discussions about the safety of AI tools have reignited. To put it simply: here’s the issue—when you use agents, skills, mcp, and similar tools, the prompts inside might be poisoned. This is not just a theoretical risk; there have already been real cases.

The core problem is a dilemma. If you turn on "danger mode," it’s definitely more exciting—the tool can fully automate control of your computer, without waiting for your confirmation each time. But what’s the cost? Once attacked, it truly helps hackers do their work automatically, leaving no room for reaction.

On the other hand, if you turn off danger mode for safety, every operation requires manual confirmation, making the process cumbersome and significantly reducing efficiency. In high-frequency trading or time-sensitive scenarios, this delay could be costly.

Ultimately, it comes down to personal judgment—choose convenience or safety. There’s no absolute answer, but at least you should be aware. Especially when dealing with crypto assets, a little extra caution is always worth it.