Show HN: Red-team LLM reasoning and agent actions (honest scoring, local-first) (github.com)

The release of "CoT Red Team Agent" v0.6.0 introduces a powerful open-source CLI and Python API designed to evaluate large language model (LLM) behaviors under adversarial conditions. This tool specifically focuses on scoring observable agent actions and generates auditable reports, moving beyond simple text assessments. By using a keyless mock provider that requires no network access, the tool provides a safe environment for developers to test LLM vulnerabilities without incurring costs. New features include the Proof-of-Action capability, which assesses agent impact through action and state transitions rather than mere refusals or generated prose. This update is significant for the AI/ML community as it enhances the safety and robustness of AI applications by providing developers with the means to conduct reproducible model attacks and detailed behavior evaluations. The CLI supports compliance scanning against OWASP's upcoming GenAI LLM Top 10 standards, making it an essential resource for developers and researchers aiming to mitigate potential risks associated with LLM deployment. The introduction of features like adaptive attacks and interactive user interfaces promises a more user-friendly experience while helping to solidify best practices in AI security testing.