dg-skill

Adversarial code review is an automated process where two sub-agents iterate by critiquing and defending code changes. This multi-agent approach continues rounds of dialogue until convergence is reached, providing multiple perspectives on software quality and security.

You can automate iterative code review by dispatching two sub-agents to critique and respond to code changes. The agents iterate through dialogue rounds until convergence, generating template-driven output to present review findings for your pull requests.

Multi-agent code review is suitable for security-sensitive changes and complex refactors. The two-agent adversarial approach provides multiple perspectives, which is beneficial when reviewing quality-critical scenarios where thorough coverage and iterative dialogue are required.

The best way to present multi-agent code review findings is through template-driven output generation. This Skill uses a comic-template HTML file to format and display the converged results from the adversarial review rounds between the two sub-agents.

Convergence-based stopping should be used when running iterative adversarial review rounds between multiple agents. It automatically halts the critique and defense dialogue once the sub-agents reach a consensus, preventing endless loops and finalizing the review output.