Show HN: Scan your AI agents for dangerous capabilities (github.com)

A new tool called MakerChecker has been announced that enables users to scan AI agents for potentially dangerous capabilities, enhancing governance and risk management in AI deployments. This tool implements a deny-by-default enforcement mechanism, requiring human approvals for high-risk actions while maintaining a cryptographically signed audit trail. It integrates seamlessly with existing frameworks like LangChain and Claude SDK, ensuring that agents can only execute actions they are explicitly granted permission for, thereby reducing the risk of unintended consequences from autonomous operations. The significance of MakerChecker lies in its ability to classify agents’ capabilities based on risk and provide a structured approach to managing them. By automatically flagging high-risk actions—such as deleting data or moving money—and requiring human oversight before execution, it addresses critical safety concerns in AI applications. Technical features include a self-hosted server with a comprehensive audit log that allows for offline verification of actions, thus enhancing trustworthiness in AI operations. With capabilities to wrap around existing tools, MakerChecker positions itself as a vital solution for organizations looking to enforce stricter controls on AI agent functions while ensuring compliance with regulatory standards.