hiddink-harness:agora

Multi-agent adversarial consensus orchestrates multiple LLMs to independently critique technical designs and debate findings until unanimous agreement, mitigating single-model blind spots and groupthink in architectural validation.

Invoke the agora skill on your design document to initiate a multi-agent review where independent LLMs compete to identify security flaws and reach evidence-based consensus on critical design risks.

Yes, both codex-exec and gemini-exec are required dependencies to enable the cross-review mechanism where multiple agents challenge and rebut each other's findings during the adversarial consensus loop.

Yes, adversarial consensus applies to complex decision-making workflows requiring high confidence by enabling peer-to-peer challenges between agents to surface hidden flaws and resolve disagreements through automated reporting.

Using a competitive multi-agent adversarial consensus loop is the best way to avoid single-model blind spots, as it forces 3+ LLMs to independently critique designs and debate findings until unanimous agreement is reached.