The Agent Skills Directory

[SAFE]: The skill provides a structured framework for multi-model adversarial code reviews. It correctly utilizes internal reference files (references/) to guide the subagents and the lead reviewer. Findings are categorized appropriately (Act On, Consider, Noted, Dismissed) to provide clear guidance to the user.
[SAFE]: The lead judgment framework explicitly instructs the lead reviewer to trace security findings and only act on those that are concrete and traceable, which is a security best practice for automated reviews.
[SAFE]: Dependency management and model selection instructions in references/conventions.md adhere to standard practices, including specific guidance on using local models (Qwen) when appropriate for low-cost tasks.

adversarial-review