reflexion:critique

Multi-agent code review evaluates solution architecture by coordinating parallel judges to independently assess completed work, debate findings, and produce a structured consensus report with actionable improvements.

The LLM-as-a-judge approach for quality assurance uses a structured evaluation framework where parallel judges independently critique work, debate findings, and build consensus to surface actionable improvements.

To validate requirements using multi-agent debate patterns, provide the artifact for critique and the system applies phase-based review steps, cross-review debates, and safety checks to produce a structured consensus report.

Yes, you can apply multi-agent consensus to check for safety and toxicity in code by including phase-based review steps and cross-review debates that evaluate code quality and produce actionable critiques.

The best way to get an actionable critique is using multi-agent debate, where parallel judges independently assess work across requirements, architecture, and code quality, then debate findings to reach supported recommendations.