Show HN: Diff two versions of an AI agent and catch silent permission changes (www.agent-kits.com)

A new tool has been introduced to enable users to compare different versions of AI agents effectively, addressing the problem of silent permission changes. Unlike traditional chatbots that produce variable outputs based on prompts, this tool employs a deterministic approach to structurally parse and analyze the two versions. It assesses each agent's Trust Level using a conservative logic from the AgentAz validator. This allows users to pinpoint exact changes, such as the loss of an “issue_refund” approval gate, and visualize the risk landscape before and after updates. The significance of this tool lies in its ability to provide clear, reproducible audits of AI agent modifications, rather than subjective interpretations. As large language models (LLMs) increasingly drive agent behavior, having a reliable way to track and validate modifications becomes essential for maintaining safety and compliance in AI systems. It operates directly within the browser without uploading or storing user data, ensuring privacy. Users can also continuously monitor for changes through Agent Watch or enforce strict version control using a GitHub Action, enhancing overall governance in AI development.