santa-method

The Santa Method skill implements a multi-agent adversarial verification framework designed to enhance the reliability and accuracy of generated outputs. It addresses the problem of single-agent bias and systematic errors that occur when one agent evaluates its own output. By employing two independent review agents with no shared context, this method ensures that content is rigorously checked before being delivered, mitigating risks associated with hallucinations, factual inaccuracies, or compliance violations.

The skill's main features include a structured, multi-phase process: generation by the primary agent, parallel independent dual review by two separate reviewers, a verdict gate that requires both reviewers to pass for the output to be considered acceptable, and a fix cycle that iterates until all issues are resolved. Each reviewer evaluates outputs against an identical rubric while remaining context-isolated, producing structured JSON assessments with clear pass/fail results, critical issues, and improvement suggestions. This ensures systematic quality control, prevents propagation of biases, and enforces convergence to a high-confidence output.

Ideal use cases include publishing content that will reach end users, ensuring regulatory or compliance adherence, shipping code to production without human oversight, verifying technical documentation, educational material, or customer-facing content, and batch content generation where systemic errors may go unnoticed. The target users are developers, technical writers, QA engineers, and content teams who require deterministic and rigorous verification for high-stakes outputs, where precision, reliability, and the avoidance of errors are paramount.