Researchers have developed a framework called TRUSS for generating software agent skills that are both functionally effective and safety reliable. The framework evaluates candidate skills against evidence from their source code and domain knowledge, as well as through controlled execution in a simulated environment. This approach has been shown to improve task performance and reduce the risk of vulnerabilities in software agents. In experiments on 168 skill artifacts, TRUSS achieved perfect precision and recall in detecting vulnerabilities, and improved task effectiveness by over 300% compared to not using skills at all. --- Why it matters: This matters because AI systems are increasingly being used in applications where safety is critical, such as healthcare and finance. The ability to generate reliable and safe agent skills can help prevent accidents and security breaches caused by flawed or malicious code. Source: https://arxiv.org/abs/2608.17588