onex-hostile-reviewer

Multi-model adversarial code review improves risk assessment by utilizing multiple models like Codex and DeepSeek-R1 to cross-check findings, which increases the accuracy of identifying and mitigating code-related risks.

To run an adversarial review on a pull request, execute the hostile-reviewer command with the specific PR number and repository, such as /hostile-reviewer --pr <N> --repo <owner/repo>, to initiate the multi-model analysis.

Yes, you need the omniintelligence review pairing library to perform adversarial code reviews and enable merge gate modes that run parallel review agents for scope, correctness, and conventions.

Static analysis performs standard code quality checks, while adversarial review combines multiple models to cross-check findings for higher accuracy in risk assessment during PR, file, or merge gate scenarios.

Yes, the tool offers both PR and file review modes to cater to different code review scenarios, allowing you to perform multi-model adversarial risk assessment on individual files.

Merge gate mode runs parallel review agents to evaluate code for scope, correctness, and conventions, providing an automated gate that ensures high-quality risk assessment before merging changes.