Show HN: Adversarial code review setup with herdr, CLI-API-proxy and Claude Code (github.com)

A new Claude Code skill has been announced that facilitates an adversarial code review process by utilizing a secondary model, GPT‑5.6 Sol, as an independent reviewer. This setup allows developers to submit code diffs or plans to Claude, which then generates a thorough review through a live interaction in a split pane powered by the herdr environment. The unique aspect of this configuration is that it not only presents the review findings but also allows users to interrogate and validate each result against the actual code. This interactive experience is designed to highlight potential flaws in the code before deployment, enhancing overall code quality and security. This advancement is significant for the AI/ML community as it demonstrates a progressive use of multiple models working in concert to improve software development practices. By integrating tools like herdr and the CLIProxyAPI to facilitate this adversarial feedback loop, developers can leverage AI capabilities not just for coding assistance, but also for rigorous testing and validation. The implementation of sandboxing via agent-safehouse further ensures a secure environment, protecting sensitive information while enabling the review process. This reflects a growing trend towards using AI in increasingly complex and safety-sensitive workflows, underscoring the potential for AI to enhance both productivity and security in software engineering.