multi-ai-review

Multi-AI consensus for code review uses a 3-stage pipeline: initial opinions, cross-review, and chairman synthesis. Claude Code, Gemini, and Codex provide separate perspectives that are weighted by concrete evidence to produce a single auditable recommendation.

Yes, AI council presets support domain auto-routing for code-review, market-regime, investment, risk assessment, and product reviews. The system detects the domain context from your prompt and automatically selects the appropriate YAML preset to tailor the workflow.

To run a multi-AI review, execute the council CLI script with your review prompt, such as `./scripts/council.sh 'Review this PR for security and architecture'`. The engine orchestrates the AI models to generate an evidence-weighted consensus.

Evidence-weighted scoring prioritizes concrete evidence, such as file and line citations, during the AI review process. It requires verification steps before any score changes are accepted, ensuring the final consensus is auditable and grounded in specific data.

Yes, YAML is a required dependency for the multi-AI review engine. The skill uses YAML formats to drive automated, reproducible AI council discussions through configurable presets and reference files.

The multi-AI council requires frontmatter name and description definitions to function, relies on external CLI tools for Claude, Gemini, and Codex, and bypasses code fences to provide actionable recommendations, meaning it is designed for structured, text-based analysis rather than direct execution.