SkillSonar：通过运行时安全技能防御Agent恶意插件攻击
A malicious agent skill can look safe when installed and turn dangerous only dur…
Agent安全是当前落地痛点，这篇论文给出了具体的运行时防御方案与量化效果，做Agent开发的同学值得参考其防御架构思路。
A malicious agent skill can look safe when installed and turn dangerous only during a real task, so this paper argues that agents need runtime safety checks, not just pre-install scanning.
恶意技能在安装时可能看似安全，仅在真实任务期间才变得危险，因此本文主张智能体需要运行时安全检查，而不仅仅是安装前的扫描。
The problem is timing: a bad skill can wait until the agent has access to useful files, tools, credentials, or external services before pushing it beyond what the user actually asked for.
问题在于时机：不良技能可以等待智能体获取到有用文件、工具、凭证或外部服务后，再将其行为推超出用户实际请求的范围。
The paper proposes SkillSonar, a safety skill that checks sensitive actions while the agent is working and decides whether to allow them, narrow them, replan, or ask the user first.
本文提出了 SkillSonar，这是一种安全技能，可在智能体工作时检查敏感操作，并决定是否允许、限制、重新规划或先征求用户同意。
On GLM-5, it cut attack success from 48.2% to 10.4% on familiar attack types and from 60.6% to 11.5% on unseen risk families.
在 GLM-5 上，它使已知攻击类型的攻击成功率从 48.2% 降至 10.4%，未知风险家族的攻击成功率从 60.6% 降至 11.5%。
A crucial result: simply installing the safety skill was much weaker.
一个关键结果：仅安装该安全技能的效果要弱得多。
The agent had to be explicitly told to consult it before acting.
必须明确指示智能体在行动前咨询该技能。
overall, the paper says scan skills before installation, check their actions during execution, and still keep hard protections like permissions and sandboxing underneath.
总体而言，本文指出应在安装前扫描技能，在执行过程中检查其操作，并在底层保留权限和沙箱等硬性保护措施。
– arxiv. org/abs/2609.01487
– arxiv.org/abs/2609.01487
Title: "Defense-as-Skill: Evolving Runtime Guard Skill for Skill-Augmented Agents"
标题：《作为技能的防御：为技能增强型智能体演进运行时防护技能》
更进一步：量化金融体系
看懂新闻只是起点——沿量化金融路径，把它变成能交付的工程能力